2 * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
19 #include "ofproto/ofproto-provider.h"
25 #include "byte-order.h"
30 #include "dynamic-string.h"
31 #include "fail-open.h"
35 #include "mac-learning.h"
36 #include "meta-flow.h"
37 #include "multipath.h"
38 #include "netdev-vport.h"
45 #include "ofp-actions.h"
46 #include "ofp-parse.h"
47 #include "ofp-print.h"
48 #include "ofproto-dpif-governor.h"
49 #include "ofproto-dpif-sflow.h"
50 #include "poll-loop.h"
55 #include "unaligned.h"
57 #include "vlan-bitmap.h"
60 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
62 COVERAGE_DEFINE(ofproto_dpif_expired);
63 COVERAGE_DEFINE(ofproto_dpif_xlate);
64 COVERAGE_DEFINE(facet_changed_rule);
65 COVERAGE_DEFINE(facet_revalidate);
66 COVERAGE_DEFINE(facet_unexpected);
67 COVERAGE_DEFINE(facet_suppress);
69 /* Maximum depth of flow table recursion (due to resubmit actions) in a
70 * flow translation. */
71 #define MAX_RESUBMIT_RECURSION 64
73 /* Number of implemented OpenFlow tables. */
74 enum { N_TABLES = 255 };
75 enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */
76 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
88 * - Do include packets and bytes from facets that have been deleted or
89 * whose own statistics have been folded into the rule.
91 * - Do include packets and bytes sent "by hand" that were accounted to
92 * the rule without any facet being involved (this is a rare corner
93 * case in rule_execute()).
95 * - Do not include packet or bytes that can be obtained from any facet's
96 * packet_count or byte_count member or that can be obtained from the
97 * datapath by, e.g., dpif_flow_get() for any subfacet.
99 uint64_t packet_count; /* Number of packets received. */
100 uint64_t byte_count; /* Number of bytes received. */
102 tag_type tag; /* Caches rule_calculate_tag() result. */
104 struct list facets; /* List of "struct facet"s. */
107 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
109 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
112 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
113 const struct flow *);
114 static struct rule_dpif *rule_dpif_lookup__(struct ofproto_dpif *,
117 static struct rule_dpif *rule_dpif_miss_rule(struct ofproto_dpif *ofproto,
118 const struct flow *flow);
120 static void rule_credit_stats(struct rule_dpif *,
121 const struct dpif_flow_stats *);
122 static void flow_push_stats(struct facet *, const struct dpif_flow_stats *);
123 static tag_type rule_calculate_tag(const struct flow *,
124 const struct minimask *, uint32_t basis);
125 static void rule_invalidate(const struct rule_dpif *);
127 #define MAX_MIRRORS 32
128 typedef uint32_t mirror_mask_t;
129 #define MIRROR_MASK_C(X) UINT32_C(X)
130 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
132 struct ofproto_dpif *ofproto; /* Owning ofproto. */
133 size_t idx; /* In ofproto's "mirrors" array. */
134 void *aux; /* Key supplied by ofproto's client. */
135 char *name; /* Identifier for log messages. */
137 /* Selection criteria. */
138 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
139 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
140 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
142 /* Output (exactly one of out == NULL and out_vlan == -1 is true). */
143 struct ofbundle *out; /* Output port or NULL. */
144 int out_vlan; /* Output VLAN or -1. */
145 mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */
148 int64_t packet_count; /* Number of packets sent. */
149 int64_t byte_count; /* Number of bytes sent. */
152 static void mirror_destroy(struct ofmirror *);
153 static void update_mirror_stats(struct ofproto_dpif *ofproto,
154 mirror_mask_t mirrors,
155 uint64_t packets, uint64_t bytes);
158 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
159 struct ofproto_dpif *ofproto; /* Owning ofproto. */
160 void *aux; /* Key supplied by ofproto's client. */
161 char *name; /* Identifier for log messages. */
164 struct list ports; /* Contains "struct ofport"s. */
165 enum port_vlan_mode vlan_mode; /* VLAN mode */
166 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
167 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
168 * NULL if all VLANs are trunked. */
169 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
170 struct bond *bond; /* Nonnull iff more than one port. */
171 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
174 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
176 /* Port mirroring info. */
177 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
178 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
179 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
182 static void bundle_remove(struct ofport *);
183 static void bundle_update(struct ofbundle *);
184 static void bundle_destroy(struct ofbundle *);
185 static void bundle_del_port(struct ofport_dpif *);
186 static void bundle_run(struct ofbundle *);
187 static void bundle_wait(struct ofbundle *);
188 static struct ofbundle *lookup_input_bundle(const struct ofproto_dpif *,
189 uint16_t in_port, bool warn,
190 struct ofport_dpif **in_ofportp);
192 /* A controller may use OFPP_NONE as the ingress port to indicate that
193 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
194 * when an input bundle is needed for validation (e.g., mirroring or
195 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
196 * any 'port' structs, so care must be taken when dealing with it. */
197 static struct ofbundle ofpp_none_bundle = {
199 .vlan_mode = PORT_VLAN_TRUNK
202 static void stp_run(struct ofproto_dpif *ofproto);
203 static void stp_wait(struct ofproto_dpif *ofproto);
204 static int set_stp_port(struct ofport *,
205 const struct ofproto_port_stp_settings *);
207 static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan);
209 struct action_xlate_ctx {
210 /* action_xlate_ctx_init() initializes these members. */
213 struct ofproto_dpif *ofproto;
215 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
216 * this flow when actions change header fields. */
219 /* stack for the push and pop actions.
220 * Each stack element is of the type "union mf_subvalue". */
222 union mf_subvalue init_stack[1024 / sizeof(union mf_subvalue)];
224 /* The packet corresponding to 'flow', or a null pointer if we are
225 * revalidating without a packet to refer to. */
226 const struct ofpbuf *packet;
228 /* Should OFPP_NORMAL update the MAC learning table? Should "learn"
229 * actions update the flow table?
231 * We want to update these tables if we are actually processing a packet,
232 * or if we are accounting for packets that the datapath has processed, but
233 * not if we are just revalidating. */
236 /* The rule that we are currently translating, or NULL. */
237 struct rule_dpif *rule;
239 /* Union of the set of TCP flags seen so far in this flow. (Used only by
240 * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules'
244 /* If nonnull, flow translation calls this function just before executing a
245 * resubmit or OFPP_TABLE action. In addition, disables logging of traces
246 * when the recursion depth is exceeded.
248 * 'rule' is the rule being submitted into. It will be null if the
249 * resubmit or OFPP_TABLE action didn't find a matching rule.
251 * This is normally null so the client has to set it manually after
252 * calling action_xlate_ctx_init(). */
253 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *rule);
255 /* If nonnull, flow translation calls this function to report some
256 * significant decision, e.g. to explain why OFPP_NORMAL translation
257 * dropped a packet. */
258 void (*report_hook)(struct action_xlate_ctx *, const char *s);
260 /* If nonnull, flow translation credits the specified statistics to each
261 * rule reached through a resubmit or OFPP_TABLE action.
263 * This is normally null so the client has to set it manually after
264 * calling action_xlate_ctx_init(). */
265 const struct dpif_flow_stats *resubmit_stats;
267 /* xlate_actions() initializes and uses these members. The client might want
268 * to look at them after it returns. */
270 struct ofpbuf *odp_actions; /* Datapath actions. */
271 tag_type tags; /* Tags associated with actions. */
272 enum slow_path_reason slow; /* 0 if fast path may be used. */
273 bool has_learn; /* Actions include NXAST_LEARN? */
274 bool has_normal; /* Actions output to OFPP_NORMAL? */
275 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
276 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
277 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
279 /* xlate_actions() initializes and uses these members, but the client has no
280 * reason to look at them. */
282 int recurse; /* Recursion level, via xlate_table_action. */
283 bool max_resubmit_trigger; /* Recursed too deeply during translation. */
284 struct flow base_flow; /* Flow at the last commit. */
285 uint32_t orig_skb_priority; /* Priority when packet arrived. */
286 uint8_t table_id; /* OpenFlow table ID where flow was found. */
287 uint32_t sflow_n_outputs; /* Number of output ports. */
288 uint32_t sflow_odp_port; /* Output port for composing sFlow action. */
289 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
290 bool exit; /* No further actions should be processed. */
293 /* Initial values of fields of the packet that may be changed during
294 * flow processing and needed later. */
295 struct initial_vals {
296 /* This is the value of vlan_tci in the packet as actually received from
297 * dpif. This is the same as the facet's flow.vlan_tci unless the packet
298 * was received via a VLAN splinter. In that case, this value is 0
299 * (because the packet as actually received from the dpif had no 802.1Q
300 * tag) but the facet's flow.vlan_tci is set to the VLAN that the splinter
303 * This member should be removed when the VLAN splinters feature is no
307 /* If received on a tunnel, the IP TOS value of the tunnel. */
308 uint8_t tunnel_ip_tos;
311 static void action_xlate_ctx_init(struct action_xlate_ctx *,
312 struct ofproto_dpif *, const struct flow *,
313 const struct initial_vals *initial_vals,
315 uint8_t tcp_flags, const struct ofpbuf *);
316 static void xlate_actions(struct action_xlate_ctx *,
317 const struct ofpact *ofpacts, size_t ofpacts_len,
318 struct ofpbuf *odp_actions);
319 static void xlate_actions_for_side_effects(struct action_xlate_ctx *,
320 const struct ofpact *ofpacts,
322 static void xlate_table_action(struct action_xlate_ctx *, uint16_t in_port,
323 uint8_t table_id, bool may_packet_in);
325 static size_t put_userspace_action(const struct ofproto_dpif *,
326 struct ofpbuf *odp_actions,
328 const union user_action_cookie *);
330 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
331 enum slow_path_reason,
332 uint64_t *stub, size_t stub_size,
333 const struct nlattr **actionsp,
334 size_t *actions_lenp);
336 static void xlate_report(struct action_xlate_ctx *ctx, const char *s);
338 /* A subfacet (see "struct subfacet" below) has three possible installation
341 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
342 * case just after the subfacet is created, just before the subfacet is
343 * destroyed, or if the datapath returns an error when we try to install a
346 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
348 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
349 * ofproto_dpif is installed in the datapath.
352 SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */
353 SF_FAST_PATH, /* Full actions are installed. */
354 SF_SLOW_PATH, /* Send-to-userspace action is installed. */
357 static const char *subfacet_path_to_string(enum subfacet_path);
359 /* A dpif flow and actions associated with a facet.
361 * See also the large comment on struct facet. */
364 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
365 struct list list_node; /* In struct facet's 'facets' list. */
366 struct facet *facet; /* Owning facet. */
368 enum odp_key_fitness key_fitness;
372 long long int used; /* Time last used; time created if not used. */
374 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
375 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
379 * These should be essentially identical for every subfacet in a facet, but
380 * may differ in trivial ways due to VLAN splinters. */
381 size_t actions_len; /* Number of bytes in actions[]. */
382 struct nlattr *actions; /* Datapath actions. */
384 enum slow_path_reason slow; /* 0 if fast path may be used. */
385 enum subfacet_path path; /* Installed in datapath? */
387 /* Initial values of the packet that may be needed later. */
388 struct initial_vals initial_vals;
390 /* Datapath port the packet arrived on. This is needed to remove
391 * flows for ports that are no longer part of the bridge. Since the
392 * flow definition only has the OpenFlow port number and the port is
393 * no longer part of the bridge, we can't determine the datapath port
394 * number needed to delete the flow from the datapath. */
395 uint32_t odp_in_port;
398 #define SUBFACET_DESTROY_MAX_BATCH 50
400 static struct subfacet *subfacet_create(struct facet *, struct flow_miss *miss,
402 static struct subfacet *subfacet_find(struct ofproto_dpif *,
403 const struct nlattr *key, size_t key_len,
405 static void subfacet_destroy(struct subfacet *);
406 static void subfacet_destroy__(struct subfacet *);
407 static void subfacet_destroy_batch(struct ofproto_dpif *,
408 struct subfacet **, int n);
409 static void subfacet_reset_dp_stats(struct subfacet *,
410 struct dpif_flow_stats *);
411 static void subfacet_update_time(struct subfacet *, long long int used);
412 static void subfacet_update_stats(struct subfacet *,
413 const struct dpif_flow_stats *);
414 static void subfacet_make_actions(struct subfacet *,
415 const struct ofpbuf *packet,
416 struct ofpbuf *odp_actions);
417 static int subfacet_install(struct subfacet *,
418 const struct nlattr *actions, size_t actions_len,
419 struct dpif_flow_stats *, enum slow_path_reason);
420 static void subfacet_uninstall(struct subfacet *);
422 static enum subfacet_path subfacet_want_path(enum slow_path_reason);
424 /* An exact-match instantiation of an OpenFlow flow.
426 * A facet associates a "struct flow", which represents the Open vSwitch
427 * userspace idea of an exact-match flow, with one or more subfacets. Each
428 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
429 * the facet. When the kernel module (or other dpif implementation) and Open
430 * vSwitch userspace agree on the definition of a flow key, there is exactly
431 * one subfacet per facet. If the dpif implementation supports more-specific
432 * flow matching than userspace, however, a facet can have more than one
433 * subfacet, each of which corresponds to some distinction in flow that
434 * userspace simply doesn't understand.
436 * Flow expiration works in terms of subfacets, so a facet must have at least
437 * one subfacet or it will never expire, leaking memory. */
440 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
441 struct list list_node; /* In owning rule's 'facets' list. */
442 struct rule_dpif *rule; /* Owning rule. */
445 struct list subfacets;
446 long long int used; /* Time last used; time created if not used. */
453 * - Do include packets and bytes sent "by hand", e.g. with
456 * - Do include packets and bytes that were obtained from the datapath
457 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
458 * DPIF_FP_ZERO_STATS).
460 * - Do not include packets or bytes that can be obtained from the
461 * datapath for any existing subfacet.
463 uint64_t packet_count; /* Number of packets received. */
464 uint64_t byte_count; /* Number of bytes received. */
466 /* Resubmit statistics. */
467 uint64_t prev_packet_count; /* Number of packets from last stats push. */
468 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
469 long long int prev_used; /* Used time from last stats push. */
472 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
473 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
474 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
476 /* Properties of datapath actions.
478 * Every subfacet has its own actions because actions can differ slightly
479 * between splintered and non-splintered subfacets due to the VLAN tag
480 * being initially different (present vs. absent). All of them have these
481 * properties in common so we just store one copy of them here. */
482 bool has_learn; /* Actions include NXAST_LEARN? */
483 bool has_normal; /* Actions output to OFPP_NORMAL? */
484 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
485 tag_type tags; /* Tags that would require revalidation. */
486 mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */
488 /* Storage for a single subfacet, to reduce malloc() time and space
489 * overhead. (A facet always has at least one subfacet and in the common
490 * case has exactly one subfacet. However, 'one_subfacet' may not
491 * always be valid, since it could have been removed after newer
492 * subfacets were pushed onto the 'subfacets' list.) */
493 struct subfacet one_subfacet;
496 static struct facet *facet_create(struct rule_dpif *,
497 const struct flow *, uint32_t hash);
498 static void facet_remove(struct facet *);
499 static void facet_free(struct facet *);
501 static struct facet *facet_find(struct ofproto_dpif *,
502 const struct flow *, uint32_t hash);
503 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
504 const struct flow *, uint32_t hash);
505 static void facet_revalidate(struct facet *);
506 static bool facet_check_consistency(struct facet *);
508 static void facet_flush_stats(struct facet *);
510 static void facet_update_time(struct facet *, long long int used);
511 static void facet_reset_counters(struct facet *);
512 static void facet_push_stats(struct facet *);
513 static void facet_learn(struct facet *);
514 static void facet_account(struct facet *);
516 static bool facet_is_controller_flow(struct facet *);
519 struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
523 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
524 struct list bundle_node; /* In struct ofbundle's "ports" list. */
525 struct cfm *cfm; /* Connectivity Fault Management, if any. */
526 tag_type tag; /* Tag associated with this port. */
527 bool may_enable; /* May be enabled in bonds. */
528 long long int carrier_seq; /* Carrier status changes. */
529 struct tnl_port *tnl_port; /* Tunnel handle, or null. */
532 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
533 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
534 long long int stp_state_entered;
536 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
538 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
540 * This is deprecated. It is only for compatibility with broken device
541 * drivers in old versions of Linux that do not properly support VLANs when
542 * VLAN devices are not used. When broken device drivers are no longer in
543 * widespread use, we will delete these interfaces. */
544 uint16_t realdev_ofp_port;
548 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
549 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
550 * traffic egressing the 'ofport' with that priority should be marked with. */
551 struct priority_to_dscp {
552 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
553 uint32_t priority; /* Priority of this queue (see struct flow). */
555 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
558 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
560 * This is deprecated. It is only for compatibility with broken device drivers
561 * in old versions of Linux that do not properly support VLANs when VLAN
562 * devices are not used. When broken device drivers are no longer in
563 * widespread use, we will delete these interfaces. */
564 struct vlan_splinter {
565 struct hmap_node realdev_vid_node;
566 struct hmap_node vlandev_node;
567 uint16_t realdev_ofp_port;
568 uint16_t vlandev_ofp_port;
572 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
573 uint32_t realdev, ovs_be16 vlan_tci);
574 static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *);
575 static void vsp_remove(struct ofport_dpif *);
576 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
578 static uint32_t ofp_port_to_odp_port(const struct ofproto_dpif *,
580 static uint16_t odp_port_to_ofp_port(const struct ofproto_dpif *,
583 static struct ofport_dpif *
584 ofport_dpif_cast(const struct ofport *ofport)
586 ovs_assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
587 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
590 static void port_run(struct ofport_dpif *);
591 static void port_run_fast(struct ofport_dpif *);
592 static void port_wait(struct ofport_dpif *);
593 static int set_cfm(struct ofport *, const struct cfm_settings *);
594 static void ofport_clear_priorities(struct ofport_dpif *);
596 struct dpif_completion {
597 struct list list_node;
598 struct ofoperation *op;
601 /* Extra information about a classifier table.
602 * Currently used just for optimized flow revalidation. */
604 /* If either of these is nonnull, then this table has a form that allows
605 * flows to be tagged to avoid revalidating most flows for the most common
606 * kinds of flow table changes. */
607 struct cls_table *catchall_table; /* Table that wildcards all fields. */
608 struct cls_table *other_table; /* Table with any other wildcard set. */
609 uint32_t basis; /* Keeps each table's tags separate. */
612 /* Reasons that we might need to revalidate every facet, and corresponding
615 * A value of 0 means that there is no need to revalidate.
617 * It would be nice to have some cleaner way to integrate with coverage
618 * counters, but with only a few reasons I guess this is good enough for
620 enum revalidate_reason {
621 REV_RECONFIGURE = 1, /* Switch configuration changed. */
622 REV_STP, /* Spanning tree protocol port status change. */
623 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
624 REV_FLOW_TABLE, /* Flow table changed. */
625 REV_INCONSISTENCY /* Facet self-check failed. */
627 COVERAGE_DEFINE(rev_reconfigure);
628 COVERAGE_DEFINE(rev_stp);
629 COVERAGE_DEFINE(rev_port_toggled);
630 COVERAGE_DEFINE(rev_flow_table);
631 COVERAGE_DEFINE(rev_inconsistency);
633 /* Drop keys are odp flow keys which have drop flows installed in the kernel.
634 * These are datapath flows which have no associated ofproto, if they did we
635 * would use facets. */
637 struct hmap_node hmap_node;
642 /* All datapaths of a given type share a single dpif backer instance. */
647 struct timer next_expiration;
648 struct hmap odp_to_ofport_map; /* ODP port to ofport mapping. */
650 struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
652 /* Facet revalidation flags applying to facets which use this backer. */
653 enum revalidate_reason need_revalidate; /* Revalidate every facet. */
654 struct tag_set revalidate_set; /* Revalidate only matching facets. */
656 struct hmap drop_keys; /* Set of dropped odp keys. */
659 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
660 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
662 static void drop_key_clear(struct dpif_backer *);
663 static struct ofport_dpif *
664 odp_port_to_ofport(const struct dpif_backer *, uint32_t odp_port);
666 struct ofproto_dpif {
667 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
669 struct dpif_backer *backer;
671 /* Special OpenFlow rules. */
672 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
673 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
679 struct netflow *netflow;
680 struct dpif_sflow *sflow;
681 struct hmap bundles; /* Contains "struct ofbundle"s. */
682 struct mac_learning *ml;
683 struct ofmirror *mirrors[MAX_MIRRORS];
685 bool has_bonded_bundles;
689 struct hmap subfacets;
690 struct governor *governor;
693 struct table_dpif tables[N_TABLES];
695 /* Support for debugging async flow mods. */
696 struct list completions;
698 bool has_bundle_action; /* True when the first bundle action appears. */
699 struct netdev_stats stats; /* To account packets generated and consumed in
704 long long int stp_last_tick;
706 /* VLAN splinters. */
707 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
708 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
711 struct sset ports; /* Set of standard port names. */
712 struct sset ghost_ports; /* Ports with no datapath port. */
713 struct sset port_poll_set; /* Queued names for port_poll() reply. */
714 int port_poll_errno; /* Last errno for port_poll() reply. */
717 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
718 * for debugging the asynchronous flow_mod implementation.) */
721 /* All existing ofproto_dpif instances, indexed by ->up.name. */
722 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
724 static void ofproto_dpif_unixctl_init(void);
726 static struct ofproto_dpif *
727 ofproto_dpif_cast(const struct ofproto *ofproto)
729 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
730 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
733 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *,
735 static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *,
737 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
738 const struct ofpbuf *,
739 const struct initial_vals *, struct ds *);
741 /* Packet processing. */
742 static void update_learning_table(struct ofproto_dpif *,
743 const struct flow *, int vlan,
746 #define FLOW_MISS_MAX_BATCH 50
747 static int handle_upcalls(struct dpif_backer *, unsigned int max_batch);
749 /* Flow expiration. */
750 static int expire(struct dpif_backer *);
753 static void send_netflow_active_timeouts(struct ofproto_dpif *);
756 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
757 static size_t compose_sflow_action(const struct ofproto_dpif *,
758 struct ofpbuf *odp_actions,
759 const struct flow *, uint32_t odp_port);
760 static void add_mirror_actions(struct action_xlate_ctx *ctx,
761 const struct flow *flow);
762 /* Global variables. */
763 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
765 /* Initial mappings of port to bridge mappings. */
766 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
768 /* Factory functions. */
771 init(const struct shash *iface_hints)
773 struct shash_node *node;
775 /* Make a local copy, since we don't own 'iface_hints' elements. */
776 SHASH_FOR_EACH(node, iface_hints) {
777 const struct iface_hint *orig_hint = node->data;
778 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
780 new_hint->br_name = xstrdup(orig_hint->br_name);
781 new_hint->br_type = xstrdup(orig_hint->br_type);
782 new_hint->ofp_port = orig_hint->ofp_port;
784 shash_add(&init_ofp_ports, node->name, new_hint);
789 enumerate_types(struct sset *types)
791 dp_enumerate_types(types);
795 enumerate_names(const char *type, struct sset *names)
797 struct ofproto_dpif *ofproto;
800 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
801 if (strcmp(type, ofproto->up.type)) {
804 sset_add(names, ofproto->up.name);
811 del(const char *type, const char *name)
816 error = dpif_open(name, type, &dpif);
818 error = dpif_delete(dpif);
825 port_open_type(const char *datapath_type, const char *port_type)
827 return dpif_port_open_type(datapath_type, port_type);
830 /* Type functions. */
832 static struct ofproto_dpif *
833 lookup_ofproto_dpif_by_port_name(const char *name)
835 struct ofproto_dpif *ofproto;
837 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
838 if (sset_contains(&ofproto->ports, name)) {
847 type_run(const char *type)
849 struct dpif_backer *backer;
853 backer = shash_find_data(&all_dpif_backers, type);
855 /* This is not necessarily a problem, since backers are only
856 * created on demand. */
860 dpif_run(backer->dpif);
862 if (backer->need_revalidate
863 || !tag_set_is_empty(&backer->revalidate_set)) {
864 struct tag_set revalidate_set = backer->revalidate_set;
865 bool need_revalidate = backer->need_revalidate;
866 struct ofproto_dpif *ofproto;
867 struct simap_node *node;
868 struct simap tmp_backers;
870 /* Handle tunnel garbage collection. */
871 simap_init(&tmp_backers);
872 simap_swap(&backer->tnl_backers, &tmp_backers);
874 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
875 struct ofport_dpif *iter;
877 if (backer != ofproto->backer) {
881 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
884 if (!iter->tnl_port) {
888 dp_port = netdev_vport_get_dpif_port(iter->up.netdev);
889 node = simap_find(&tmp_backers, dp_port);
891 simap_put(&backer->tnl_backers, dp_port, node->data);
892 simap_delete(&tmp_backers, node);
893 node = simap_find(&backer->tnl_backers, dp_port);
895 node = simap_find(&backer->tnl_backers, dp_port);
897 uint32_t odp_port = UINT32_MAX;
899 if (!dpif_port_add(backer->dpif, iter->up.netdev,
901 simap_put(&backer->tnl_backers, dp_port, odp_port);
902 node = simap_find(&backer->tnl_backers, dp_port);
907 iter->odp_port = node ? node->data : OVSP_NONE;
908 if (tnl_port_reconfigure(&iter->up, iter->odp_port,
910 backer->need_revalidate = REV_RECONFIGURE;
915 SIMAP_FOR_EACH (node, &tmp_backers) {
916 dpif_port_del(backer->dpif, node->data);
918 simap_destroy(&tmp_backers);
920 switch (backer->need_revalidate) {
921 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
922 case REV_STP: COVERAGE_INC(rev_stp); break;
923 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
924 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
925 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
928 if (backer->need_revalidate) {
929 /* Clear the drop_keys in case we should now be accepting some
930 * formerly dropped flows. */
931 drop_key_clear(backer);
934 /* Clear the revalidation flags. */
935 tag_set_init(&backer->revalidate_set);
936 backer->need_revalidate = 0;
938 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
939 struct facet *facet, *next;
941 if (ofproto->backer != backer) {
945 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
947 || tag_set_intersects(&revalidate_set, facet->tags)) {
948 facet_revalidate(facet);
954 if (timer_expired(&backer->next_expiration)) {
955 int delay = expire(backer);
956 timer_set_duration(&backer->next_expiration, delay);
959 /* Check for port changes in the dpif. */
960 while ((error = dpif_port_poll(backer->dpif, &devname)) == 0) {
961 struct ofproto_dpif *ofproto;
962 struct dpif_port port;
964 /* Don't report on the datapath's device. */
965 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
969 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
970 &all_ofproto_dpifs) {
971 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
976 ofproto = lookup_ofproto_dpif_by_port_name(devname);
977 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
978 /* The port was removed. If we know the datapath,
979 * report it through poll_set(). If we don't, it may be
980 * notifying us of a removal we initiated, so ignore it.
981 * If there's a pending ENOBUFS, let it stand, since
982 * everything will be reevaluated. */
983 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
984 sset_add(&ofproto->port_poll_set, devname);
985 ofproto->port_poll_errno = 0;
987 } else if (!ofproto) {
988 /* The port was added, but we don't know with which
989 * ofproto we should associate it. Delete it. */
990 dpif_port_del(backer->dpif, port.port_no);
992 dpif_port_destroy(&port);
998 if (error != EAGAIN) {
999 struct ofproto_dpif *ofproto;
1001 /* There was some sort of error, so propagate it to all
1002 * ofprotos that use this backer. */
1003 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
1004 &all_ofproto_dpifs) {
1005 if (ofproto->backer == backer) {
1006 sset_clear(&ofproto->port_poll_set);
1007 ofproto->port_poll_errno = error;
1016 type_run_fast(const char *type)
1018 struct dpif_backer *backer;
1021 backer = shash_find_data(&all_dpif_backers, type);
1023 /* This is not necessarily a problem, since backers are only
1024 * created on demand. */
1028 /* Handle one or more batches of upcalls, until there's nothing left to do
1029 * or until we do a fixed total amount of work.
1031 * We do work in batches because it can be much cheaper to set up a number
1032 * of flows and fire off their patches all at once. We do multiple batches
1033 * because in some cases handling a packet can cause another packet to be
1034 * queued almost immediately as part of the return flow. Both
1035 * optimizations can make major improvements on some benchmarks and
1036 * presumably for real traffic as well. */
1038 while (work < FLOW_MISS_MAX_BATCH) {
1039 int retval = handle_upcalls(backer, FLOW_MISS_MAX_BATCH - work);
1050 type_wait(const char *type)
1052 struct dpif_backer *backer;
1054 backer = shash_find_data(&all_dpif_backers, type);
1056 /* This is not necessarily a problem, since backers are only
1057 * created on demand. */
1061 timer_wait(&backer->next_expiration);
1064 /* Basic life-cycle. */
1066 static int add_internal_flows(struct ofproto_dpif *);
1068 static struct ofproto *
1071 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1072 return &ofproto->up;
1076 dealloc(struct ofproto *ofproto_)
1078 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1083 close_dpif_backer(struct dpif_backer *backer)
1085 struct shash_node *node;
1087 ovs_assert(backer->refcount > 0);
1089 if (--backer->refcount) {
1093 drop_key_clear(backer);
1094 hmap_destroy(&backer->drop_keys);
1096 simap_destroy(&backer->tnl_backers);
1097 hmap_destroy(&backer->odp_to_ofport_map);
1098 node = shash_find(&all_dpif_backers, backer->type);
1100 shash_delete(&all_dpif_backers, node);
1101 dpif_close(backer->dpif);
1106 /* Datapath port slated for removal from datapath. */
1107 struct odp_garbage {
1108 struct list list_node;
1113 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1115 struct dpif_backer *backer;
1116 struct dpif_port_dump port_dump;
1117 struct dpif_port port;
1118 struct shash_node *node;
1119 struct list garbage_list;
1120 struct odp_garbage *garbage, *next;
1126 backer = shash_find_data(&all_dpif_backers, type);
1133 backer_name = xasprintf("ovs-%s", type);
1135 /* Remove any existing datapaths, since we assume we're the only
1136 * userspace controlling the datapath. */
1138 dp_enumerate_names(type, &names);
1139 SSET_FOR_EACH(name, &names) {
1140 struct dpif *old_dpif;
1142 /* Don't remove our backer if it exists. */
1143 if (!strcmp(name, backer_name)) {
1147 if (dpif_open(name, type, &old_dpif)) {
1148 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1150 dpif_delete(old_dpif);
1151 dpif_close(old_dpif);
1154 sset_destroy(&names);
1156 backer = xmalloc(sizeof *backer);
1158 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1161 VLOG_ERR("failed to open datapath of type %s: %s", type,
1167 backer->type = xstrdup(type);
1168 backer->refcount = 1;
1169 hmap_init(&backer->odp_to_ofport_map);
1170 hmap_init(&backer->drop_keys);
1171 timer_set_duration(&backer->next_expiration, 1000);
1172 backer->need_revalidate = 0;
1173 simap_init(&backer->tnl_backers);
1174 tag_set_init(&backer->revalidate_set);
1177 dpif_flow_flush(backer->dpif);
1179 /* Loop through the ports already on the datapath and remove any
1180 * that we don't need anymore. */
1181 list_init(&garbage_list);
1182 dpif_port_dump_start(&port_dump, backer->dpif);
1183 while (dpif_port_dump_next(&port_dump, &port)) {
1184 node = shash_find(&init_ofp_ports, port.name);
1185 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1186 garbage = xmalloc(sizeof *garbage);
1187 garbage->odp_port = port.port_no;
1188 list_push_front(&garbage_list, &garbage->list_node);
1191 dpif_port_dump_done(&port_dump);
1193 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1194 dpif_port_del(backer->dpif, garbage->odp_port);
1195 list_remove(&garbage->list_node);
1199 shash_add(&all_dpif_backers, type, backer);
1201 error = dpif_recv_set(backer->dpif, true);
1203 VLOG_ERR("failed to listen on datapath of type %s: %s",
1204 type, strerror(error));
1205 close_dpif_backer(backer);
1213 construct(struct ofproto *ofproto_)
1215 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1216 struct shash_node *node, *next;
1221 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1226 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1227 ofproto_init_max_ports(ofproto_, MIN(max_ports, OFPP_MAX));
1229 ofproto->n_matches = 0;
1231 ofproto->netflow = NULL;
1232 ofproto->sflow = NULL;
1233 ofproto->stp = NULL;
1234 hmap_init(&ofproto->bundles);
1235 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1236 for (i = 0; i < MAX_MIRRORS; i++) {
1237 ofproto->mirrors[i] = NULL;
1239 ofproto->has_bonded_bundles = false;
1241 hmap_init(&ofproto->facets);
1242 hmap_init(&ofproto->subfacets);
1243 ofproto->governor = NULL;
1245 for (i = 0; i < N_TABLES; i++) {
1246 struct table_dpif *table = &ofproto->tables[i];
1248 table->catchall_table = NULL;
1249 table->other_table = NULL;
1250 table->basis = random_uint32();
1253 list_init(&ofproto->completions);
1255 ofproto_dpif_unixctl_init();
1257 ofproto->has_mirrors = false;
1258 ofproto->has_bundle_action = false;
1260 hmap_init(&ofproto->vlandev_map);
1261 hmap_init(&ofproto->realdev_vid_map);
1263 sset_init(&ofproto->ports);
1264 sset_init(&ofproto->ghost_ports);
1265 sset_init(&ofproto->port_poll_set);
1266 ofproto->port_poll_errno = 0;
1268 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1269 struct iface_hint *iface_hint = node->data;
1271 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1272 /* Check if the datapath already has this port. */
1273 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1274 sset_add(&ofproto->ports, node->name);
1277 free(iface_hint->br_name);
1278 free(iface_hint->br_type);
1280 shash_delete(&init_ofp_ports, node);
1284 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1285 hash_string(ofproto->up.name, 0));
1286 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1288 ofproto_init_tables(ofproto_, N_TABLES);
1289 error = add_internal_flows(ofproto);
1290 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1296 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1297 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1299 struct ofputil_flow_mod fm;
1302 match_init_catchall(&fm.match);
1304 match_set_reg(&fm.match, 0, id);
1305 fm.new_cookie = htonll(0);
1306 fm.cookie = htonll(0);
1307 fm.cookie_mask = htonll(0);
1308 fm.table_id = TBL_INTERNAL;
1309 fm.command = OFPFC_ADD;
1310 fm.idle_timeout = 0;
1311 fm.hard_timeout = 0;
1315 fm.ofpacts = ofpacts->data;
1316 fm.ofpacts_len = ofpacts->size;
1318 error = ofproto_flow_mod(&ofproto->up, &fm);
1320 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1321 id, ofperr_to_string(error));
1325 *rulep = rule_dpif_lookup__(ofproto, &fm.match.flow, TBL_INTERNAL);
1326 ovs_assert(*rulep != NULL);
1332 add_internal_flows(struct ofproto_dpif *ofproto)
1334 struct ofpact_controller *controller;
1335 uint64_t ofpacts_stub[128 / 8];
1336 struct ofpbuf ofpacts;
1340 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1343 controller = ofpact_put_CONTROLLER(&ofpacts);
1344 controller->max_len = UINT16_MAX;
1345 controller->controller_id = 0;
1346 controller->reason = OFPR_NO_MATCH;
1347 ofpact_pad(&ofpacts);
1349 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1354 ofpbuf_clear(&ofpacts);
1355 error = add_internal_flow(ofproto, id++, &ofpacts,
1356 &ofproto->no_packet_in_rule);
1361 complete_operations(struct ofproto_dpif *ofproto)
1363 struct dpif_completion *c, *next;
1365 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1366 ofoperation_complete(c->op, 0);
1367 list_remove(&c->list_node);
1373 destruct(struct ofproto *ofproto_)
1375 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1376 struct rule_dpif *rule, *next_rule;
1377 struct oftable *table;
1380 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1381 complete_operations(ofproto);
1383 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1384 struct cls_cursor cursor;
1386 cls_cursor_init(&cursor, &table->cls, NULL);
1387 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1388 ofproto_rule_destroy(&rule->up);
1392 for (i = 0; i < MAX_MIRRORS; i++) {
1393 mirror_destroy(ofproto->mirrors[i]);
1396 netflow_destroy(ofproto->netflow);
1397 dpif_sflow_destroy(ofproto->sflow);
1398 hmap_destroy(&ofproto->bundles);
1399 mac_learning_destroy(ofproto->ml);
1401 hmap_destroy(&ofproto->facets);
1402 hmap_destroy(&ofproto->subfacets);
1403 governor_destroy(ofproto->governor);
1405 hmap_destroy(&ofproto->vlandev_map);
1406 hmap_destroy(&ofproto->realdev_vid_map);
1408 sset_destroy(&ofproto->ports);
1409 sset_destroy(&ofproto->ghost_ports);
1410 sset_destroy(&ofproto->port_poll_set);
1412 close_dpif_backer(ofproto->backer);
1416 run_fast(struct ofproto *ofproto_)
1418 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1419 struct ofport_dpif *ofport;
1421 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1422 port_run_fast(ofport);
1429 run(struct ofproto *ofproto_)
1431 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1432 struct ofport_dpif *ofport;
1433 struct ofbundle *bundle;
1437 complete_operations(ofproto);
1440 error = run_fast(ofproto_);
1445 if (ofproto->netflow) {
1446 if (netflow_run(ofproto->netflow)) {
1447 send_netflow_active_timeouts(ofproto);
1450 if (ofproto->sflow) {
1451 dpif_sflow_run(ofproto->sflow);
1454 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1457 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1462 mac_learning_run(ofproto->ml, &ofproto->backer->revalidate_set);
1464 /* Check the consistency of a random facet, to aid debugging. */
1465 if (!hmap_is_empty(&ofproto->facets)
1466 && !ofproto->backer->need_revalidate) {
1467 struct facet *facet;
1469 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
1470 struct facet, hmap_node);
1471 if (!tag_set_intersects(&ofproto->backer->revalidate_set,
1473 if (!facet_check_consistency(facet)) {
1474 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1479 if (ofproto->governor) {
1482 governor_run(ofproto->governor);
1484 /* If the governor has shrunk to its minimum size and the number of
1485 * subfacets has dwindled, then drop the governor entirely.
1487 * For hysteresis, the number of subfacets to drop the governor is
1488 * smaller than the number needed to trigger its creation. */
1489 n_subfacets = hmap_count(&ofproto->subfacets);
1490 if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold
1491 && governor_is_idle(ofproto->governor)) {
1492 governor_destroy(ofproto->governor);
1493 ofproto->governor = NULL;
1501 wait(struct ofproto *ofproto_)
1503 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1504 struct ofport_dpif *ofport;
1505 struct ofbundle *bundle;
1507 if (!clogged && !list_is_empty(&ofproto->completions)) {
1508 poll_immediate_wake();
1511 dpif_wait(ofproto->backer->dpif);
1512 dpif_recv_wait(ofproto->backer->dpif);
1513 if (ofproto->sflow) {
1514 dpif_sflow_wait(ofproto->sflow);
1516 if (!tag_set_is_empty(&ofproto->backer->revalidate_set)) {
1517 poll_immediate_wake();
1519 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1522 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1523 bundle_wait(bundle);
1525 if (ofproto->netflow) {
1526 netflow_wait(ofproto->netflow);
1528 mac_learning_wait(ofproto->ml);
1530 if (ofproto->backer->need_revalidate) {
1531 /* Shouldn't happen, but if it does just go around again. */
1532 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1533 poll_immediate_wake();
1535 if (ofproto->governor) {
1536 governor_wait(ofproto->governor);
1541 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1543 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1545 simap_increase(usage, "facets", hmap_count(&ofproto->facets));
1546 simap_increase(usage, "subfacets", hmap_count(&ofproto->subfacets));
1550 flush(struct ofproto *ofproto_)
1552 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1553 struct subfacet *subfacet, *next_subfacet;
1554 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1558 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1559 &ofproto->subfacets) {
1560 if (subfacet->path != SF_NOT_INSTALLED) {
1561 batch[n_batch++] = subfacet;
1562 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1563 subfacet_destroy_batch(ofproto, batch, n_batch);
1567 subfacet_destroy(subfacet);
1572 subfacet_destroy_batch(ofproto, batch, n_batch);
1577 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1578 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1580 *arp_match_ip = true;
1581 *actions = (OFPUTIL_A_OUTPUT |
1582 OFPUTIL_A_SET_VLAN_VID |
1583 OFPUTIL_A_SET_VLAN_PCP |
1584 OFPUTIL_A_STRIP_VLAN |
1585 OFPUTIL_A_SET_DL_SRC |
1586 OFPUTIL_A_SET_DL_DST |
1587 OFPUTIL_A_SET_NW_SRC |
1588 OFPUTIL_A_SET_NW_DST |
1589 OFPUTIL_A_SET_NW_TOS |
1590 OFPUTIL_A_SET_TP_SRC |
1591 OFPUTIL_A_SET_TP_DST |
1596 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1598 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1599 struct dpif_dp_stats s;
1601 strcpy(ots->name, "classifier");
1603 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1605 ots->lookup_count = htonll(s.n_hit + s.n_missed);
1606 ots->matched_count = htonll(s.n_hit + ofproto->n_matches);
1609 static struct ofport *
1612 struct ofport_dpif *port = xmalloc(sizeof *port);
1617 port_dealloc(struct ofport *port_)
1619 struct ofport_dpif *port = ofport_dpif_cast(port_);
1624 port_construct(struct ofport *port_)
1626 struct ofport_dpif *port = ofport_dpif_cast(port_);
1627 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1628 const struct netdev *netdev = port->up.netdev;
1629 struct dpif_port dpif_port;
1632 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1633 port->bundle = NULL;
1635 port->tag = tag_create_random();
1636 port->may_enable = true;
1637 port->stp_port = NULL;
1638 port->stp_state = STP_DISABLED;
1639 port->tnl_port = NULL;
1640 hmap_init(&port->priorities);
1641 port->realdev_ofp_port = 0;
1642 port->vlandev_vid = 0;
1643 port->carrier_seq = netdev_get_carrier_resets(netdev);
1645 if (netdev_vport_is_patch(netdev)) {
1646 /* XXX By bailing out here, we don't do required sFlow work. */
1647 port->odp_port = OVSP_NONE;
1651 error = dpif_port_query_by_name(ofproto->backer->dpif,
1652 netdev_vport_get_dpif_port(netdev),
1658 port->odp_port = dpif_port.port_no;
1660 if (netdev_get_tunnel_config(netdev)) {
1661 port->tnl_port = tnl_port_add(&port->up, port->odp_port);
1663 /* Sanity-check that a mapping doesn't already exist. This
1664 * shouldn't happen for non-tunnel ports. */
1665 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1666 VLOG_ERR("port %s already has an OpenFlow port number",
1668 dpif_port_destroy(&dpif_port);
1672 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1673 hash_int(port->odp_port, 0));
1675 dpif_port_destroy(&dpif_port);
1677 if (ofproto->sflow) {
1678 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1685 port_destruct(struct ofport *port_)
1687 struct ofport_dpif *port = ofport_dpif_cast(port_);
1688 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1689 const char *dp_port_name = netdev_vport_get_dpif_port(port->up.netdev);
1690 const char *devname = netdev_get_name(port->up.netdev);
1692 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1693 /* The underlying device is still there, so delete it. This
1694 * happens when the ofproto is being destroyed, since the caller
1695 * assumes that removal of attached ports will happen as part of
1697 if (!port->tnl_port) {
1698 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1700 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1703 if (port->odp_port != OVSP_NONE && !port->tnl_port) {
1704 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1707 tnl_port_del(port->tnl_port);
1708 sset_find_and_delete(&ofproto->ports, devname);
1709 sset_find_and_delete(&ofproto->ghost_ports, devname);
1710 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1711 bundle_remove(port_);
1712 set_cfm(port_, NULL);
1713 if (ofproto->sflow) {
1714 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1717 ofport_clear_priorities(port);
1718 hmap_destroy(&port->priorities);
1722 port_modified(struct ofport *port_)
1724 struct ofport_dpif *port = ofport_dpif_cast(port_);
1726 if (port->bundle && port->bundle->bond) {
1727 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1732 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1734 struct ofport_dpif *port = ofport_dpif_cast(port_);
1735 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1736 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1738 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1739 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1740 OFPUTIL_PC_NO_PACKET_IN)) {
1741 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1743 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1744 bundle_update(port->bundle);
1750 set_sflow(struct ofproto *ofproto_,
1751 const struct ofproto_sflow_options *sflow_options)
1753 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1754 struct dpif_sflow *ds = ofproto->sflow;
1756 if (sflow_options) {
1758 struct ofport_dpif *ofport;
1760 ds = ofproto->sflow = dpif_sflow_create();
1761 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1762 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1764 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1766 dpif_sflow_set_options(ds, sflow_options);
1769 dpif_sflow_destroy(ds);
1770 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1771 ofproto->sflow = NULL;
1778 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1780 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1787 struct ofproto_dpif *ofproto;
1789 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1790 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1791 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1794 if (cfm_configure(ofport->cfm, s)) {
1800 cfm_destroy(ofport->cfm);
1806 get_cfm_status(const struct ofport *ofport_,
1807 struct ofproto_cfm_status *status)
1809 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1812 status->faults = cfm_get_fault(ofport->cfm);
1813 status->remote_opstate = cfm_get_opup(ofport->cfm);
1814 status->health = cfm_get_health(ofport->cfm);
1815 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
1822 /* Spanning Tree. */
1825 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1827 struct ofproto_dpif *ofproto = ofproto_;
1828 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1829 struct ofport_dpif *ofport;
1831 ofport = stp_port_get_aux(sp);
1833 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1834 ofproto->up.name, port_num);
1836 struct eth_header *eth = pkt->l2;
1838 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1839 if (eth_addr_is_zero(eth->eth_src)) {
1840 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1841 "with unknown MAC", ofproto->up.name, port_num);
1843 send_packet(ofport, pkt);
1849 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1851 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1853 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1855 /* Only revalidate flows if the configuration changed. */
1856 if (!s != !ofproto->stp) {
1857 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1861 if (!ofproto->stp) {
1862 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1863 send_bpdu_cb, ofproto);
1864 ofproto->stp_last_tick = time_msec();
1867 stp_set_bridge_id(ofproto->stp, s->system_id);
1868 stp_set_bridge_priority(ofproto->stp, s->priority);
1869 stp_set_hello_time(ofproto->stp, s->hello_time);
1870 stp_set_max_age(ofproto->stp, s->max_age);
1871 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1873 struct ofport *ofport;
1875 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1876 set_stp_port(ofport, NULL);
1879 stp_destroy(ofproto->stp);
1880 ofproto->stp = NULL;
1887 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1889 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1893 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1894 s->designated_root = stp_get_designated_root(ofproto->stp);
1895 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1904 update_stp_port_state(struct ofport_dpif *ofport)
1906 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1907 enum stp_state state;
1909 /* Figure out new state. */
1910 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1914 if (ofport->stp_state != state) {
1915 enum ofputil_port_state of_state;
1918 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1919 netdev_get_name(ofport->up.netdev),
1920 stp_state_name(ofport->stp_state),
1921 stp_state_name(state));
1922 if (stp_learn_in_state(ofport->stp_state)
1923 != stp_learn_in_state(state)) {
1924 /* xxx Learning action flows should also be flushed. */
1925 mac_learning_flush(ofproto->ml,
1926 &ofproto->backer->revalidate_set);
1928 fwd_change = stp_forward_in_state(ofport->stp_state)
1929 != stp_forward_in_state(state);
1931 ofproto->backer->need_revalidate = REV_STP;
1932 ofport->stp_state = state;
1933 ofport->stp_state_entered = time_msec();
1935 if (fwd_change && ofport->bundle) {
1936 bundle_update(ofport->bundle);
1939 /* Update the STP state bits in the OpenFlow port description. */
1940 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
1941 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
1942 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
1943 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
1944 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
1946 ofproto_port_set_state(&ofport->up, of_state);
1950 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1951 * caller is responsible for assigning STP port numbers and ensuring
1952 * there are no duplicates. */
1954 set_stp_port(struct ofport *ofport_,
1955 const struct ofproto_port_stp_settings *s)
1957 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1958 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1959 struct stp_port *sp = ofport->stp_port;
1961 if (!s || !s->enable) {
1963 ofport->stp_port = NULL;
1964 stp_port_disable(sp);
1965 update_stp_port_state(ofport);
1968 } else if (sp && stp_port_no(sp) != s->port_num
1969 && ofport == stp_port_get_aux(sp)) {
1970 /* The port-id changed, so disable the old one if it's not
1971 * already in use by another port. */
1972 stp_port_disable(sp);
1975 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1976 stp_port_enable(sp);
1978 stp_port_set_aux(sp, ofport);
1979 stp_port_set_priority(sp, s->priority);
1980 stp_port_set_path_cost(sp, s->path_cost);
1982 update_stp_port_state(ofport);
1988 get_stp_port_status(struct ofport *ofport_,
1989 struct ofproto_port_stp_status *s)
1991 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1992 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1993 struct stp_port *sp = ofport->stp_port;
1995 if (!ofproto->stp || !sp) {
2001 s->port_id = stp_port_get_id(sp);
2002 s->state = stp_port_get_state(sp);
2003 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2004 s->role = stp_port_get_role(sp);
2005 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2011 stp_run(struct ofproto_dpif *ofproto)
2014 long long int now = time_msec();
2015 long long int elapsed = now - ofproto->stp_last_tick;
2016 struct stp_port *sp;
2019 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2020 ofproto->stp_last_tick = now;
2022 while (stp_get_changed_port(ofproto->stp, &sp)) {
2023 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2026 update_stp_port_state(ofport);
2030 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2031 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2037 stp_wait(struct ofproto_dpif *ofproto)
2040 poll_timer_wait(1000);
2044 /* Returns true if STP should process 'flow'. */
2046 stp_should_process_flow(const struct flow *flow)
2048 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
2052 stp_process_packet(const struct ofport_dpif *ofport,
2053 const struct ofpbuf *packet)
2055 struct ofpbuf payload = *packet;
2056 struct eth_header *eth = payload.data;
2057 struct stp_port *sp = ofport->stp_port;
2059 /* Sink packets on ports that have STP disabled when the bridge has
2061 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
2065 /* Trim off padding on payload. */
2066 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
2067 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
2070 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
2071 stp_received_bpdu(sp, payload.data, payload.size);
2075 static struct priority_to_dscp *
2076 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
2078 struct priority_to_dscp *pdscp;
2081 hash = hash_int(priority, 0);
2082 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
2083 if (pdscp->priority == priority) {
2091 ofport_clear_priorities(struct ofport_dpif *ofport)
2093 struct priority_to_dscp *pdscp, *next;
2095 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
2096 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2102 set_queues(struct ofport *ofport_,
2103 const struct ofproto_port_queue *qdscp_list,
2106 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2107 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2108 struct hmap new = HMAP_INITIALIZER(&new);
2111 for (i = 0; i < n_qdscp; i++) {
2112 struct priority_to_dscp *pdscp;
2116 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
2117 if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue,
2122 pdscp = get_priority(ofport, priority);
2124 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2126 pdscp = xmalloc(sizeof *pdscp);
2127 pdscp->priority = priority;
2129 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2132 if (pdscp->dscp != dscp) {
2134 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2137 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
2140 if (!hmap_is_empty(&ofport->priorities)) {
2141 ofport_clear_priorities(ofport);
2142 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2145 hmap_swap(&new, &ofport->priorities);
2153 /* Expires all MAC learning entries associated with 'bundle' and forces its
2154 * ofproto to revalidate every flow.
2156 * Normally MAC learning entries are removed only from the ofproto associated
2157 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2158 * are removed from every ofproto. When patch ports and SLB bonds are in use
2159 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2160 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2161 * with the host from which it migrated. */
2163 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2165 struct ofproto_dpif *ofproto = bundle->ofproto;
2166 struct mac_learning *ml = ofproto->ml;
2167 struct mac_entry *mac, *next_mac;
2169 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2170 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2171 if (mac->port.p == bundle) {
2173 struct ofproto_dpif *o;
2175 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2177 struct mac_entry *e;
2179 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
2182 mac_learning_expire(o->ml, e);
2188 mac_learning_expire(ml, mac);
2193 static struct ofbundle *
2194 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2196 struct ofbundle *bundle;
2198 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2199 &ofproto->bundles) {
2200 if (bundle->aux == aux) {
2207 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
2208 * ones that are found to 'bundles'. */
2210 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
2211 void **auxes, size_t n_auxes,
2212 struct hmapx *bundles)
2216 hmapx_init(bundles);
2217 for (i = 0; i < n_auxes; i++) {
2218 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
2220 hmapx_add(bundles, bundle);
2226 bundle_update(struct ofbundle *bundle)
2228 struct ofport_dpif *port;
2230 bundle->floodable = true;
2231 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2232 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2233 || !stp_forward_in_state(port->stp_state)) {
2234 bundle->floodable = false;
2241 bundle_del_port(struct ofport_dpif *port)
2243 struct ofbundle *bundle = port->bundle;
2245 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2247 list_remove(&port->bundle_node);
2248 port->bundle = NULL;
2251 lacp_slave_unregister(bundle->lacp, port);
2254 bond_slave_unregister(bundle->bond, port);
2257 bundle_update(bundle);
2261 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
2262 struct lacp_slave_settings *lacp)
2264 struct ofport_dpif *port;
2266 port = get_ofp_port(bundle->ofproto, ofp_port);
2271 if (port->bundle != bundle) {
2272 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2274 bundle_del_port(port);
2277 port->bundle = bundle;
2278 list_push_back(&bundle->ports, &port->bundle_node);
2279 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2280 || !stp_forward_in_state(port->stp_state)) {
2281 bundle->floodable = false;
2285 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2286 lacp_slave_register(bundle->lacp, port, lacp);
2293 bundle_destroy(struct ofbundle *bundle)
2295 struct ofproto_dpif *ofproto;
2296 struct ofport_dpif *port, *next_port;
2303 ofproto = bundle->ofproto;
2304 for (i = 0; i < MAX_MIRRORS; i++) {
2305 struct ofmirror *m = ofproto->mirrors[i];
2307 if (m->out == bundle) {
2309 } else if (hmapx_find_and_delete(&m->srcs, bundle)
2310 || hmapx_find_and_delete(&m->dsts, bundle)) {
2311 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2316 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2317 bundle_del_port(port);
2320 bundle_flush_macs(bundle, true);
2321 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2323 free(bundle->trunks);
2324 lacp_destroy(bundle->lacp);
2325 bond_destroy(bundle->bond);
2330 bundle_set(struct ofproto *ofproto_, void *aux,
2331 const struct ofproto_bundle_settings *s)
2333 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2334 bool need_flush = false;
2335 struct ofport_dpif *port;
2336 struct ofbundle *bundle;
2337 unsigned long *trunks;
2343 bundle_destroy(bundle_lookup(ofproto, aux));
2347 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2348 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2350 bundle = bundle_lookup(ofproto, aux);
2352 bundle = xmalloc(sizeof *bundle);
2354 bundle->ofproto = ofproto;
2355 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2356 hash_pointer(aux, 0));
2358 bundle->name = NULL;
2360 list_init(&bundle->ports);
2361 bundle->vlan_mode = PORT_VLAN_TRUNK;
2363 bundle->trunks = NULL;
2364 bundle->use_priority_tags = s->use_priority_tags;
2365 bundle->lacp = NULL;
2366 bundle->bond = NULL;
2368 bundle->floodable = true;
2370 bundle->src_mirrors = 0;
2371 bundle->dst_mirrors = 0;
2372 bundle->mirror_out = 0;
2375 if (!bundle->name || strcmp(s->name, bundle->name)) {
2377 bundle->name = xstrdup(s->name);
2382 if (!bundle->lacp) {
2383 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2384 bundle->lacp = lacp_create();
2386 lacp_configure(bundle->lacp, s->lacp);
2388 lacp_destroy(bundle->lacp);
2389 bundle->lacp = NULL;
2392 /* Update set of ports. */
2394 for (i = 0; i < s->n_slaves; i++) {
2395 if (!bundle_add_port(bundle, s->slaves[i],
2396 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2400 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2401 struct ofport_dpif *next_port;
2403 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2404 for (i = 0; i < s->n_slaves; i++) {
2405 if (s->slaves[i] == port->up.ofp_port) {
2410 bundle_del_port(port);
2414 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2416 if (list_is_empty(&bundle->ports)) {
2417 bundle_destroy(bundle);
2421 /* Set VLAN tagging mode */
2422 if (s->vlan_mode != bundle->vlan_mode
2423 || s->use_priority_tags != bundle->use_priority_tags) {
2424 bundle->vlan_mode = s->vlan_mode;
2425 bundle->use_priority_tags = s->use_priority_tags;
2430 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2431 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2433 if (vlan != bundle->vlan) {
2434 bundle->vlan = vlan;
2438 /* Get trunked VLANs. */
2439 switch (s->vlan_mode) {
2440 case PORT_VLAN_ACCESS:
2444 case PORT_VLAN_TRUNK:
2445 trunks = CONST_CAST(unsigned long *, s->trunks);
2448 case PORT_VLAN_NATIVE_UNTAGGED:
2449 case PORT_VLAN_NATIVE_TAGGED:
2450 if (vlan != 0 && (!s->trunks
2451 || !bitmap_is_set(s->trunks, vlan)
2452 || bitmap_is_set(s->trunks, 0))) {
2453 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2455 trunks = bitmap_clone(s->trunks, 4096);
2457 trunks = bitmap_allocate1(4096);
2459 bitmap_set1(trunks, vlan);
2460 bitmap_set0(trunks, 0);
2462 trunks = CONST_CAST(unsigned long *, s->trunks);
2469 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2470 free(bundle->trunks);
2471 if (trunks == s->trunks) {
2472 bundle->trunks = vlan_bitmap_clone(trunks);
2474 bundle->trunks = trunks;
2479 if (trunks != s->trunks) {
2484 if (!list_is_short(&bundle->ports)) {
2485 bundle->ofproto->has_bonded_bundles = true;
2487 if (bond_reconfigure(bundle->bond, s->bond)) {
2488 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2491 bundle->bond = bond_create(s->bond);
2492 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2495 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2496 bond_slave_register(bundle->bond, port, port->up.netdev);
2499 bond_destroy(bundle->bond);
2500 bundle->bond = NULL;
2503 /* If we changed something that would affect MAC learning, un-learn
2504 * everything on this port and force flow revalidation. */
2506 bundle_flush_macs(bundle, false);
2513 bundle_remove(struct ofport *port_)
2515 struct ofport_dpif *port = ofport_dpif_cast(port_);
2516 struct ofbundle *bundle = port->bundle;
2519 bundle_del_port(port);
2520 if (list_is_empty(&bundle->ports)) {
2521 bundle_destroy(bundle);
2522 } else if (list_is_short(&bundle->ports)) {
2523 bond_destroy(bundle->bond);
2524 bundle->bond = NULL;
2530 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2532 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2533 struct ofport_dpif *port = port_;
2534 uint8_t ea[ETH_ADDR_LEN];
2537 error = netdev_get_etheraddr(port->up.netdev, ea);
2539 struct ofpbuf packet;
2542 ofpbuf_init(&packet, 0);
2543 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2545 memcpy(packet_pdu, pdu, pdu_size);
2547 send_packet(port, &packet);
2548 ofpbuf_uninit(&packet);
2550 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2551 "%s (%s)", port->bundle->name,
2552 netdev_get_name(port->up.netdev), strerror(error));
2557 bundle_send_learning_packets(struct ofbundle *bundle)
2559 struct ofproto_dpif *ofproto = bundle->ofproto;
2560 int error, n_packets, n_errors;
2561 struct mac_entry *e;
2563 error = n_packets = n_errors = 0;
2564 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2565 if (e->port.p != bundle) {
2566 struct ofpbuf *learning_packet;
2567 struct ofport_dpif *port;
2571 /* The assignment to "port" is unnecessary but makes "grep"ing for
2572 * struct ofport_dpif more effective. */
2573 learning_packet = bond_compose_learning_packet(bundle->bond,
2577 ret = send_packet(port, learning_packet);
2578 ofpbuf_delete(learning_packet);
2588 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2589 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2590 "packets, last error was: %s",
2591 bundle->name, n_errors, n_packets, strerror(error));
2593 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2594 bundle->name, n_packets);
2599 bundle_run(struct ofbundle *bundle)
2602 lacp_run(bundle->lacp, send_pdu_cb);
2605 struct ofport_dpif *port;
2607 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2608 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2611 bond_run(bundle->bond, &bundle->ofproto->backer->revalidate_set,
2612 lacp_status(bundle->lacp));
2613 if (bond_should_send_learning_packets(bundle->bond)) {
2614 bundle_send_learning_packets(bundle);
2620 bundle_wait(struct ofbundle *bundle)
2623 lacp_wait(bundle->lacp);
2626 bond_wait(bundle->bond);
2633 mirror_scan(struct ofproto_dpif *ofproto)
2637 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2638 if (!ofproto->mirrors[idx]) {
2645 static struct ofmirror *
2646 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2650 for (i = 0; i < MAX_MIRRORS; i++) {
2651 struct ofmirror *mirror = ofproto->mirrors[i];
2652 if (mirror && mirror->aux == aux) {
2660 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2662 mirror_update_dups(struct ofproto_dpif *ofproto)
2666 for (i = 0; i < MAX_MIRRORS; i++) {
2667 struct ofmirror *m = ofproto->mirrors[i];
2670 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2674 for (i = 0; i < MAX_MIRRORS; i++) {
2675 struct ofmirror *m1 = ofproto->mirrors[i];
2682 for (j = i + 1; j < MAX_MIRRORS; j++) {
2683 struct ofmirror *m2 = ofproto->mirrors[j];
2685 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2686 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2687 m2->dup_mirrors |= m1->dup_mirrors;
2694 mirror_set(struct ofproto *ofproto_, void *aux,
2695 const struct ofproto_mirror_settings *s)
2697 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2698 mirror_mask_t mirror_bit;
2699 struct ofbundle *bundle;
2700 struct ofmirror *mirror;
2701 struct ofbundle *out;
2702 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2703 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2706 mirror = mirror_lookup(ofproto, aux);
2708 mirror_destroy(mirror);
2714 idx = mirror_scan(ofproto);
2716 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2718 ofproto->up.name, MAX_MIRRORS, s->name);
2722 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2723 mirror->ofproto = ofproto;
2726 mirror->out_vlan = -1;
2727 mirror->name = NULL;
2730 if (!mirror->name || strcmp(s->name, mirror->name)) {
2732 mirror->name = xstrdup(s->name);
2735 /* Get the new configuration. */
2736 if (s->out_bundle) {
2737 out = bundle_lookup(ofproto, s->out_bundle);
2739 mirror_destroy(mirror);
2745 out_vlan = s->out_vlan;
2747 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2748 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2750 /* If the configuration has not changed, do nothing. */
2751 if (hmapx_equals(&srcs, &mirror->srcs)
2752 && hmapx_equals(&dsts, &mirror->dsts)
2753 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2754 && mirror->out == out
2755 && mirror->out_vlan == out_vlan)
2757 hmapx_destroy(&srcs);
2758 hmapx_destroy(&dsts);
2762 hmapx_swap(&srcs, &mirror->srcs);
2763 hmapx_destroy(&srcs);
2765 hmapx_swap(&dsts, &mirror->dsts);
2766 hmapx_destroy(&dsts);
2768 free(mirror->vlans);
2769 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2772 mirror->out_vlan = out_vlan;
2774 /* Update bundles. */
2775 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2776 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2777 if (hmapx_contains(&mirror->srcs, bundle)) {
2778 bundle->src_mirrors |= mirror_bit;
2780 bundle->src_mirrors &= ~mirror_bit;
2783 if (hmapx_contains(&mirror->dsts, bundle)) {
2784 bundle->dst_mirrors |= mirror_bit;
2786 bundle->dst_mirrors &= ~mirror_bit;
2789 if (mirror->out == bundle) {
2790 bundle->mirror_out |= mirror_bit;
2792 bundle->mirror_out &= ~mirror_bit;
2796 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2797 ofproto->has_mirrors = true;
2798 mac_learning_flush(ofproto->ml,
2799 &ofproto->backer->revalidate_set);
2800 mirror_update_dups(ofproto);
2806 mirror_destroy(struct ofmirror *mirror)
2808 struct ofproto_dpif *ofproto;
2809 mirror_mask_t mirror_bit;
2810 struct ofbundle *bundle;
2817 ofproto = mirror->ofproto;
2818 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2819 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2821 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2822 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2823 bundle->src_mirrors &= ~mirror_bit;
2824 bundle->dst_mirrors &= ~mirror_bit;
2825 bundle->mirror_out &= ~mirror_bit;
2828 hmapx_destroy(&mirror->srcs);
2829 hmapx_destroy(&mirror->dsts);
2830 free(mirror->vlans);
2832 ofproto->mirrors[mirror->idx] = NULL;
2836 mirror_update_dups(ofproto);
2838 ofproto->has_mirrors = false;
2839 for (i = 0; i < MAX_MIRRORS; i++) {
2840 if (ofproto->mirrors[i]) {
2841 ofproto->has_mirrors = true;
2848 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2849 uint64_t *packets, uint64_t *bytes)
2851 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2852 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2855 *packets = *bytes = UINT64_MAX;
2859 *packets = mirror->packet_count;
2860 *bytes = mirror->byte_count;
2866 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2868 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2869 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2870 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2876 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2878 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2879 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2880 return bundle && bundle->mirror_out != 0;
2884 forward_bpdu_changed(struct ofproto *ofproto_)
2886 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2887 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2891 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2894 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2895 mac_learning_set_idle_time(ofproto->ml, idle_time);
2896 mac_learning_set_max_entries(ofproto->ml, max_entries);
2901 static struct ofport_dpif *
2902 get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
2904 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2905 return ofport ? ofport_dpif_cast(ofport) : NULL;
2908 static struct ofport_dpif *
2909 get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
2911 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
2912 return port && &ofproto->up == port->up.ofproto ? port : NULL;
2916 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2917 struct ofproto_port *ofproto_port,
2918 struct dpif_port *dpif_port)
2920 ofproto_port->name = dpif_port->name;
2921 ofproto_port->type = dpif_port->type;
2922 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2925 static struct ofport_dpif *
2926 ofport_get_peer(const struct ofport_dpif *ofport_dpif)
2928 const struct ofproto_dpif *ofproto;
2931 peer = netdev_vport_patch_peer(ofport_dpif->up.netdev);
2936 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2937 struct ofport *ofport;
2939 ofport = shash_find_data(&ofproto->up.port_by_name, peer);
2940 if (ofport && ofport->ofproto->ofproto_class == &ofproto_dpif_class) {
2941 return ofport_dpif_cast(ofport);
2948 port_run_fast(struct ofport_dpif *ofport)
2950 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2951 struct ofpbuf packet;
2953 ofpbuf_init(&packet, 0);
2954 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2955 send_packet(ofport, &packet);
2956 ofpbuf_uninit(&packet);
2961 port_run(struct ofport_dpif *ofport)
2963 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2964 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2965 bool enable = netdev_get_carrier(ofport->up.netdev);
2967 ofport->carrier_seq = carrier_seq;
2969 port_run_fast(ofport);
2971 if (ofport->tnl_port
2972 && tnl_port_reconfigure(&ofport->up, ofport->odp_port,
2973 &ofport->tnl_port)) {
2974 ofproto_dpif_cast(ofport->up.ofproto)->backer->need_revalidate = true;
2978 int cfm_opup = cfm_get_opup(ofport->cfm);
2980 cfm_run(ofport->cfm);
2981 enable = enable && !cfm_get_fault(ofport->cfm);
2983 if (cfm_opup >= 0) {
2984 enable = enable && cfm_opup;
2988 if (ofport->bundle) {
2989 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2990 if (carrier_changed) {
2991 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2995 if (ofport->may_enable != enable) {
2996 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2998 if (ofproto->has_bundle_action) {
2999 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
3003 ofport->may_enable = enable;
3007 port_wait(struct ofport_dpif *ofport)
3010 cfm_wait(ofport->cfm);
3015 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3016 struct ofproto_port *ofproto_port)
3018 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3019 struct dpif_port dpif_port;
3022 if (sset_contains(&ofproto->ghost_ports, devname)) {
3023 const char *type = netdev_get_type_from_name(devname);
3025 /* We may be called before ofproto->up.port_by_name is populated with
3026 * the appropriate ofport. For this reason, we must get the name and
3027 * type from the netdev layer directly. */
3029 const struct ofport *ofport;
3031 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3032 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3033 ofproto_port->name = xstrdup(devname);
3034 ofproto_port->type = xstrdup(type);
3040 if (!sset_contains(&ofproto->ports, devname)) {
3043 error = dpif_port_query_by_name(ofproto->backer->dpif,
3044 devname, &dpif_port);
3046 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3052 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3054 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3055 const char *dp_port_name = netdev_vport_get_dpif_port(netdev);
3056 const char *devname = netdev_get_name(netdev);
3058 if (netdev_vport_is_patch(netdev)) {
3059 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3063 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3064 uint32_t port_no = UINT32_MAX;
3067 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3071 if (netdev_get_tunnel_config(netdev)) {
3072 simap_put(&ofproto->backer->tnl_backers, dp_port_name, port_no);
3076 if (netdev_get_tunnel_config(netdev)) {
3077 sset_add(&ofproto->ghost_ports, devname);
3079 sset_add(&ofproto->ports, devname);
3085 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
3087 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3088 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3095 sset_find_and_delete(&ofproto->ghost_ports,
3096 netdev_get_name(ofport->up.netdev));
3097 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3098 if (!ofport->tnl_port) {
3099 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3101 /* The caller is going to close ofport->up.netdev. If this is a
3102 * bonded port, then the bond is using that netdev, so remove it
3103 * from the bond. The client will need to reconfigure everything
3104 * after deleting ports, so then the slave will get re-added. */
3105 bundle_remove(&ofport->up);
3112 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3114 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3117 error = netdev_get_stats(ofport->up.netdev, stats);
3119 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3120 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3122 /* ofproto->stats.tx_packets represents packets that we created
3123 * internally and sent to some port (e.g. packets sent with
3124 * send_packet()). Account for them as if they had come from
3125 * OFPP_LOCAL and got forwarded. */
3127 if (stats->rx_packets != UINT64_MAX) {
3128 stats->rx_packets += ofproto->stats.tx_packets;
3131 if (stats->rx_bytes != UINT64_MAX) {
3132 stats->rx_bytes += ofproto->stats.tx_bytes;
3135 /* ofproto->stats.rx_packets represents packets that were received on
3136 * some port and we processed internally and dropped (e.g. STP).
3137 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3139 if (stats->tx_packets != UINT64_MAX) {
3140 stats->tx_packets += ofproto->stats.rx_packets;
3143 if (stats->tx_bytes != UINT64_MAX) {
3144 stats->tx_bytes += ofproto->stats.rx_bytes;
3151 /* Account packets for LOCAL port. */
3153 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
3154 size_t tx_size, size_t rx_size)
3156 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3159 ofproto->stats.rx_packets++;
3160 ofproto->stats.rx_bytes += rx_size;
3163 ofproto->stats.tx_packets++;
3164 ofproto->stats.tx_bytes += tx_size;
3168 struct port_dump_state {
3173 struct ofproto_port port;
3178 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3180 *statep = xzalloc(sizeof(struct port_dump_state));
3185 port_dump_next(const struct ofproto *ofproto_, void *state_,
3186 struct ofproto_port *port)
3188 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3189 struct port_dump_state *state = state_;
3190 const struct sset *sset;
3191 struct sset_node *node;
3193 if (state->has_port) {
3194 ofproto_port_destroy(&state->port);
3195 state->has_port = false;
3197 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3198 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3201 error = port_query_by_name(ofproto_, node->name, &state->port);
3203 *port = state->port;
3204 state->has_port = true;
3206 } else if (error != ENODEV) {
3211 if (!state->ghost) {
3212 state->ghost = true;
3215 return port_dump_next(ofproto_, state_, port);
3222 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3224 struct port_dump_state *state = state_;
3226 if (state->has_port) {
3227 ofproto_port_destroy(&state->port);
3234 port_poll(const struct ofproto *ofproto_, char **devnamep)
3236 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3238 if (ofproto->port_poll_errno) {
3239 int error = ofproto->port_poll_errno;
3240 ofproto->port_poll_errno = 0;
3244 if (sset_is_empty(&ofproto->port_poll_set)) {
3248 *devnamep = sset_pop(&ofproto->port_poll_set);
3253 port_poll_wait(const struct ofproto *ofproto_)
3255 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3256 dpif_port_poll_wait(ofproto->backer->dpif);
3260 port_is_lacp_current(const struct ofport *ofport_)
3262 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3263 return (ofport->bundle && ofport->bundle->lacp
3264 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3268 /* Upcall handling. */
3270 /* Flow miss batching.
3272 * Some dpifs implement operations faster when you hand them off in a batch.
3273 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3274 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3275 * more packets, plus possibly installing the flow in the dpif.
3277 * So far we only batch the operations that affect flow setup time the most.
3278 * It's possible to batch more than that, but the benefit might be minimal. */
3280 struct hmap_node hmap_node;
3281 struct ofproto_dpif *ofproto;
3283 enum odp_key_fitness key_fitness;
3284 const struct nlattr *key;
3286 struct initial_vals initial_vals;
3287 struct list packets;
3288 enum dpif_upcall_type upcall_type;
3289 uint32_t odp_in_port;
3292 struct flow_miss_op {
3293 struct dpif_op dpif_op;
3294 void *garbage; /* Pointer to pass to free(), NULL if none. */
3295 uint64_t stub[1024 / 8]; /* Temporary buffer. */
3298 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3299 * OpenFlow controller as necessary according to their individual
3300 * configurations. */
3302 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3303 const struct flow *flow)
3305 struct ofputil_packet_in pin;
3307 pin.packet = packet->data;
3308 pin.packet_len = packet->size;
3309 pin.reason = OFPR_NO_MATCH;
3310 pin.controller_id = 0;
3315 pin.send_len = 0; /* not used for flow table misses */
3317 flow_get_metadata(flow, &pin.fmd);
3319 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3322 static enum slow_path_reason
3323 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
3324 const struct ofport_dpif *ofport, const struct ofpbuf *packet)
3328 } else if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
3330 cfm_process_heartbeat(ofport->cfm, packet);
3333 } else if (ofport->bundle && ofport->bundle->lacp
3334 && flow->dl_type == htons(ETH_TYPE_LACP)) {
3336 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
3339 } else if (ofproto->stp && stp_should_process_flow(flow)) {
3341 stp_process_packet(ofport, packet);
3349 static struct flow_miss *
3350 flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto,
3351 const struct flow *flow, uint32_t hash)
3353 struct flow_miss *miss;
3355 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3356 if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) {
3364 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3365 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3366 * 'miss' is associated with a subfacet the caller must also initialize the
3367 * returned op->subfacet, and if anything needs to be freed after processing
3368 * the op, the caller must initialize op->garbage also. */
3370 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3371 struct flow_miss_op *op)
3373 if (miss->flow.vlan_tci != miss->initial_vals.vlan_tci) {
3374 /* This packet was received on a VLAN splinter port. We
3375 * added a VLAN to the packet to make the packet resemble
3376 * the flow, but the actions were composed assuming that
3377 * the packet contained no VLAN. So, we must remove the
3378 * VLAN header from the packet before trying to execute the
3380 eth_pop_vlan(packet);
3384 op->dpif_op.type = DPIF_OP_EXECUTE;
3385 op->dpif_op.u.execute.key = miss->key;
3386 op->dpif_op.u.execute.key_len = miss->key_len;
3387 op->dpif_op.u.execute.packet = packet;
3390 /* Helper for handle_flow_miss_without_facet() and
3391 * handle_flow_miss_with_facet(). */
3393 handle_flow_miss_common(struct rule_dpif *rule,
3394 struct ofpbuf *packet, const struct flow *flow)
3396 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3398 ofproto->n_matches++;
3400 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
3402 * Extra-special case for fail-open mode.
3404 * We are in fail-open mode and the packet matched the fail-open
3405 * rule, but we are connected to a controller too. We should send
3406 * the packet up to the controller in the hope that it will try to
3407 * set up a flow and thereby allow us to exit fail-open.
3409 * See the top-level comment in fail-open.c for more information.
3411 send_packet_in_miss(ofproto, packet, flow);
3415 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3416 * 'miss', is likely to be worth tracking in detail in userspace and (usually)
3417 * installing a datapath flow. The answer is usually "yes" (a return value of
3418 * true). However, for short flows the cost of bookkeeping is much higher than
3419 * the benefits, so when the datapath holds a large number of flows we impose
3420 * some heuristics to decide which flows are likely to be worth tracking. */
3422 flow_miss_should_make_facet(struct ofproto_dpif *ofproto,
3423 struct flow_miss *miss, uint32_t hash)
3425 if (!ofproto->governor) {
3428 n_subfacets = hmap_count(&ofproto->subfacets);
3429 if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) {
3433 ofproto->governor = governor_create(ofproto->up.name);
3436 return governor_should_install_flow(ofproto->governor, hash,
3437 list_size(&miss->packets));
3440 /* Handles 'miss', which matches 'rule', without creating a facet or subfacet
3441 * or creating any datapath flow. May add an "execute" operation to 'ops' and
3442 * increment '*n_ops'. */
3444 handle_flow_miss_without_facet(struct flow_miss *miss,
3445 struct rule_dpif *rule,
3446 struct flow_miss_op *ops, size_t *n_ops)
3448 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3449 long long int now = time_msec();
3450 struct action_xlate_ctx ctx;
3451 struct ofpbuf *packet;
3453 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3454 struct flow_miss_op *op = &ops[*n_ops];
3455 struct dpif_flow_stats stats;
3456 struct ofpbuf odp_actions;
3458 COVERAGE_INC(facet_suppress);
3460 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
3462 dpif_flow_stats_extract(&miss->flow, packet, now, &stats);
3463 rule_credit_stats(rule, &stats);
3465 action_xlate_ctx_init(&ctx, ofproto, &miss->flow,
3466 &miss->initial_vals, rule, 0, packet);
3467 ctx.resubmit_stats = &stats;
3468 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
3471 if (odp_actions.size) {
3472 struct dpif_execute *execute = &op->dpif_op.u.execute;
3474 init_flow_miss_execute_op(miss, packet, op);
3475 execute->actions = odp_actions.data;
3476 execute->actions_len = odp_actions.size;
3477 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
3481 ofpbuf_uninit(&odp_actions);
3486 /* Handles 'miss', which matches 'facet'. May add any required datapath
3487 * operations to 'ops', incrementing '*n_ops' for each new op.
3489 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3490 * This is really important only for new facets: if we just called time_msec()
3491 * here, then the new subfacet or its packets could look (occasionally) as
3492 * though it was used some time after the facet was used. That can make a
3493 * one-packet flow look like it has a nonzero duration, which looks odd in
3494 * e.g. NetFlow statistics. */
3496 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3498 struct flow_miss_op *ops, size_t *n_ops)
3500 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3501 enum subfacet_path want_path;
3502 struct subfacet *subfacet;
3503 struct ofpbuf *packet;
3505 subfacet = subfacet_create(facet, miss, now);
3507 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3508 struct flow_miss_op *op = &ops[*n_ops];
3509 struct dpif_flow_stats stats;
3510 struct ofpbuf odp_actions;
3512 handle_flow_miss_common(facet->rule, packet, &miss->flow);
3514 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
3515 if (!subfacet->actions || subfacet->slow) {
3516 subfacet_make_actions(subfacet, packet, &odp_actions);
3519 dpif_flow_stats_extract(&facet->flow, packet, now, &stats);
3520 subfacet_update_stats(subfacet, &stats);
3522 if (subfacet->actions_len) {
3523 struct dpif_execute *execute = &op->dpif_op.u.execute;
3525 init_flow_miss_execute_op(miss, packet, op);
3526 if (!subfacet->slow) {
3527 execute->actions = subfacet->actions;
3528 execute->actions_len = subfacet->actions_len;
3529 ofpbuf_uninit(&odp_actions);
3531 execute->actions = odp_actions.data;
3532 execute->actions_len = odp_actions.size;
3533 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
3538 ofpbuf_uninit(&odp_actions);
3542 want_path = subfacet_want_path(subfacet->slow);
3543 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
3544 struct flow_miss_op *op = &ops[(*n_ops)++];
3545 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3547 subfacet->path = want_path;
3550 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3551 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3552 put->key = miss->key;
3553 put->key_len = miss->key_len;
3554 if (want_path == SF_FAST_PATH) {
3555 put->actions = subfacet->actions;
3556 put->actions_len = subfacet->actions_len;
3558 compose_slow_path(ofproto, &facet->flow, subfacet->slow,
3559 op->stub, sizeof op->stub,
3560 &put->actions, &put->actions_len);
3566 /* Handles flow miss 'miss'. May add any required datapath operations
3567 * to 'ops', incrementing '*n_ops' for each new op. */
3569 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3572 struct ofproto_dpif *ofproto = miss->ofproto;
3573 struct facet *facet;
3577 /* The caller must ensure that miss->hmap_node.hash contains
3578 * flow_hash(miss->flow, 0). */
3579 hash = miss->hmap_node.hash;
3581 facet = facet_lookup_valid(ofproto, &miss->flow, hash);
3583 struct rule_dpif *rule = rule_dpif_lookup(ofproto, &miss->flow);
3585 if (!flow_miss_should_make_facet(ofproto, miss, hash)) {
3586 handle_flow_miss_without_facet(miss, rule, ops, n_ops);
3590 facet = facet_create(rule, &miss->flow, hash);
3595 handle_flow_miss_with_facet(miss, facet, now, ops, n_ops);
3598 static struct drop_key *
3599 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3602 struct drop_key *drop_key;
3604 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3605 &backer->drop_keys) {
3606 if (drop_key->key_len == key_len
3607 && !memcmp(drop_key->key, key, key_len)) {
3615 drop_key_clear(struct dpif_backer *backer)
3617 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3618 struct drop_key *drop_key, *next;
3620 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3623 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3625 if (error && !VLOG_DROP_WARN(&rl)) {
3626 struct ds ds = DS_EMPTY_INITIALIZER;
3627 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3628 VLOG_WARN("Failed to delete drop key (%s) (%s)", strerror(error),
3633 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3634 free(drop_key->key);
3639 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
3640 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
3641 * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as
3642 * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto'
3643 * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that
3644 * 'packet' ingressed.
3646 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
3647 * 'flow''s in_port to OFPP_NONE.
3649 * This function does post-processing on data returned from
3650 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
3651 * of the upcall processing logic. In particular, if the extracted in_port is
3652 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
3653 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
3654 * a VLAN header onto 'packet' (if it is nonnull).
3656 * Optionally, if 'initial_vals' is nonnull, sets 'initial_vals->vlan_tci'
3657 * to the VLAN TCI with which the packet was really received, that is, the
3658 * actual VLAN TCI extracted by odp_flow_key_to_flow(). (This differs from
3659 * the value returned in flow->vlan_tci only for packets received on
3660 * VLAN splinters.) Also, if received on an IP tunnel, sets
3661 * 'initial_vals->tunnel_ip_tos' to the tunnel's IP TOS.
3663 * Similarly, this function also includes some logic to help with tunnels. It
3664 * may modify 'flow' as necessary to make the tunneling implementation
3665 * transparent to the upcall processing logic.
3667 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
3668 * or some other positive errno if there are other problems. */
3670 ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
3671 const struct nlattr *key, size_t key_len,
3672 struct flow *flow, enum odp_key_fitness *fitnessp,
3673 struct ofproto_dpif **ofproto, uint32_t *odp_in_port,
3674 struct initial_vals *initial_vals)
3676 const struct ofport_dpif *port;
3677 enum odp_key_fitness fitness;
3680 fitness = odp_flow_key_to_flow(key, key_len, flow);
3681 if (fitness == ODP_FIT_ERROR) {
3687 initial_vals->vlan_tci = flow->vlan_tci;
3688 initial_vals->tunnel_ip_tos = flow->tunnel.ip_tos;
3692 *odp_in_port = flow->in_port;
3695 if (tnl_port_should_receive(flow)) {
3696 const struct ofport *ofport = tnl_port_receive(flow);
3698 flow->in_port = OFPP_NONE;
3701 port = ofport_dpif_cast(ofport);
3703 /* We can't reproduce 'key' from 'flow'. */
3704 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3706 /* XXX: Since the tunnel module is not scoped per backer, it's
3707 * theoretically possible that we'll receive an ofport belonging to an
3708 * entirely different datapath. In practice, this can't happen because
3709 * no platforms has two separate datapaths which each support
3711 ovs_assert(ofproto_dpif_cast(port->up.ofproto)->backer == backer);
3713 port = odp_port_to_ofport(backer, flow->in_port);
3715 flow->in_port = OFPP_NONE;
3719 flow->in_port = port->up.ofp_port;
3720 if (vsp_adjust_flow(ofproto_dpif_cast(port->up.ofproto), flow)) {
3722 /* Make the packet resemble the flow, so that it gets sent to
3723 * an OpenFlow controller properly, so that it looks correct
3724 * for sFlow, and so that flow_extract() will get the correct
3725 * vlan_tci if it is called on 'packet'.
3727 * The allocated space inside 'packet' probably also contains
3728 * 'key', that is, both 'packet' and 'key' are probably part of
3729 * a struct dpif_upcall (see the large comment on that
3730 * structure definition), so pushing data on 'packet' is in
3731 * general not a good idea since it could overwrite 'key' or
3732 * free it as a side effect. However, it's OK in this special
3733 * case because we know that 'packet' is inside a Netlink
3734 * attribute: pushing 4 bytes will just overwrite the 4-byte
3735 * "struct nlattr", which is fine since we don't need that
3736 * header anymore. */
3737 eth_push_vlan(packet, flow->vlan_tci);
3739 /* We can't reproduce 'key' from 'flow'. */
3740 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3746 *ofproto = ofproto_dpif_cast(port->up.ofproto);
3751 *fitnessp = fitness;
3757 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
3760 struct dpif_upcall *upcall;
3761 struct flow_miss *miss;
3762 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3763 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3764 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3774 /* Construct the to-do list.
3776 * This just amounts to extracting the flow from each packet and sticking
3777 * the packets that have the same flow in the same "flow_miss" structure so
3778 * that we can process them together. */
3781 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3782 struct flow_miss *miss = &misses[n_misses];
3783 struct flow_miss *existing_miss;
3784 struct ofproto_dpif *ofproto;
3785 uint32_t odp_in_port;
3790 error = ofproto_receive(backer, upcall->packet, upcall->key,
3791 upcall->key_len, &flow, &miss->key_fitness,
3792 &ofproto, &odp_in_port, &miss->initial_vals);
3793 if (error == ENODEV) {
3794 struct drop_key *drop_key;
3796 /* Received packet on port for which we couldn't associate
3797 * an ofproto. This can happen if a port is removed while
3798 * traffic is being received. Print a rate-limited message
3799 * in case it happens frequently. Install a drop flow so
3800 * that future packets of the flow are inexpensively dropped
3802 VLOG_INFO_RL(&rl, "received packet on unassociated port %"PRIu32,
3805 drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len);
3807 drop_key = xmalloc(sizeof *drop_key);
3808 drop_key->key = xmemdup(upcall->key, upcall->key_len);
3809 drop_key->key_len = upcall->key_len;
3811 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3812 hash_bytes(drop_key->key, drop_key->key_len, 0));
3813 dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
3814 drop_key->key, drop_key->key_len, NULL, 0, NULL);
3821 flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
3822 &flow.tunnel, flow.in_port, &miss->flow);
3824 /* Add other packets to a to-do list. */
3825 hash = flow_hash(&miss->flow, 0);
3826 existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash);
3827 if (!existing_miss) {
3828 hmap_insert(&todo, &miss->hmap_node, hash);
3829 miss->ofproto = ofproto;
3830 miss->key = upcall->key;
3831 miss->key_len = upcall->key_len;
3832 miss->upcall_type = upcall->type;
3833 miss->odp_in_port = odp_in_port;
3834 list_init(&miss->packets);
3838 miss = existing_miss;
3840 list_push_back(&miss->packets, &upcall->packet->list_node);
3843 /* Process each element in the to-do list, constructing the set of
3844 * operations to batch. */
3846 HMAP_FOR_EACH (miss, hmap_node, &todo) {
3847 handle_flow_miss(miss, flow_miss_ops, &n_ops);
3849 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3851 /* Execute batch. */
3852 for (i = 0; i < n_ops; i++) {
3853 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3855 dpif_operate(backer->dpif, dpif_ops, n_ops);
3858 for (i = 0; i < n_ops; i++) {
3859 free(flow_miss_ops[i].garbage);
3861 hmap_destroy(&todo);
3864 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL }
3865 classify_upcall(const struct dpif_upcall *upcall)
3867 union user_action_cookie cookie;
3869 /* First look at the upcall type. */
3870 switch (upcall->type) {
3871 case DPIF_UC_ACTION:
3877 case DPIF_N_UC_TYPES:
3879 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
3883 /* "action" upcalls need a closer look. */
3884 if (!upcall->userdata) {
3885 VLOG_WARN_RL(&rl, "action upcall missing cookie");
3888 if (nl_attr_get_size(upcall->userdata) != sizeof(cookie)) {
3889 VLOG_WARN_RL(&rl, "action upcall cookie has unexpected size %zu",
3890 nl_attr_get_size(upcall->userdata));
3893 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof(cookie));
3894 switch (cookie.type) {
3895 case USER_ACTION_COOKIE_SFLOW:
3896 return SFLOW_UPCALL;
3898 case USER_ACTION_COOKIE_SLOW_PATH:
3901 case USER_ACTION_COOKIE_UNSPEC:
3903 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64,
3904 nl_attr_get_u64(upcall->userdata));
3910 handle_sflow_upcall(struct dpif_backer *backer,
3911 const struct dpif_upcall *upcall)
3913 struct ofproto_dpif *ofproto;
3914 union user_action_cookie cookie;
3916 uint32_t odp_in_port;
3918 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3919 &flow, NULL, &ofproto, &odp_in_port, NULL)
3920 || !ofproto->sflow) {
3924 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof(cookie));
3925 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
3926 odp_in_port, &cookie);
3930 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
3932 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
3933 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
3934 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
3939 ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH);
3942 for (n_processed = 0; n_processed < max_batch; n_processed++) {
3943 struct dpif_upcall *upcall = &misses[n_misses];
3944 struct ofpbuf *buf = &miss_bufs[n_misses];
3947 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
3948 sizeof miss_buf_stubs[n_misses]);
3949 error = dpif_recv(backer->dpif, upcall, buf);
3955 switch (classify_upcall(upcall)) {
3957 /* Handle it later. */
3962 handle_sflow_upcall(backer, upcall);
3972 /* Handle deferred MISS_UPCALL processing. */
3973 handle_miss_upcalls(backer, misses, n_misses);
3974 for (i = 0; i < n_misses; i++) {
3975 ofpbuf_uninit(&miss_bufs[i]);
3981 /* Flow expiration. */
3983 static int subfacet_max_idle(const struct ofproto_dpif *);
3984 static void update_stats(struct dpif_backer *);
3985 static void rule_expire(struct rule_dpif *);
3986 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
3988 /* This function is called periodically by run(). Its job is to collect
3989 * updates for the flows that have been installed into the datapath, most
3990 * importantly when they last were used, and then use that information to
3991 * expire flows that have not been used recently.
3993 * Returns the number of milliseconds after which it should be called again. */
3995 expire(struct dpif_backer *backer)
3997 struct ofproto_dpif *ofproto;
3998 int max_idle = INT32_MAX;
4000 /* Periodically clear out the drop keys in an effort to keep them
4001 * relatively few. */
4002 drop_key_clear(backer);
4004 /* Update stats for each flow in the backer. */
4005 update_stats(backer);
4007 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4008 struct rule *rule, *next_rule;
4011 if (ofproto->backer != backer) {
4015 /* Expire subfacets that have been idle too long. */
4016 dp_max_idle = subfacet_max_idle(ofproto);
4017 expire_subfacets(ofproto, dp_max_idle);
4019 max_idle = MIN(max_idle, dp_max_idle);
4021 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
4023 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
4024 &ofproto->up.expirable) {
4025 rule_expire(rule_dpif_cast(rule));
4028 /* All outstanding data in existing flows has been accounted, so it's a
4029 * good time to do bond rebalancing. */
4030 if (ofproto->has_bonded_bundles) {
4031 struct ofbundle *bundle;
4033 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
4035 bond_rebalance(bundle->bond, &backer->revalidate_set);
4041 return MIN(max_idle, 1000);
4044 /* Updates flow table statistics given that the datapath just reported 'stats'
4045 * as 'subfacet''s statistics. */
4047 update_subfacet_stats(struct subfacet *subfacet,
4048 const struct dpif_flow_stats *stats)
4050 struct facet *facet = subfacet->facet;
4052 if (stats->n_packets >= subfacet->dp_packet_count) {
4053 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
4054 facet->packet_count += extra;
4056 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4059 if (stats->n_bytes >= subfacet->dp_byte_count) {
4060 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
4062 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4065 subfacet->dp_packet_count = stats->n_packets;
4066 subfacet->dp_byte_count = stats->n_bytes;
4068 facet->tcp_flags |= stats->tcp_flags;
4070 subfacet_update_time(subfacet, stats->used);
4071 if (facet->accounted_bytes < facet->byte_count) {
4073 facet_account(facet);
4074 facet->accounted_bytes = facet->byte_count;
4076 facet_push_stats(facet);
4079 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
4080 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
4082 delete_unexpected_flow(struct ofproto_dpif *ofproto,
4083 const struct nlattr *key, size_t key_len)
4085 if (!VLOG_DROP_WARN(&rl)) {
4089 odp_flow_key_format(key, key_len, &s);
4090 VLOG_WARN("unexpected flow on %s: %s", ofproto->up.name, ds_cstr(&s));
4094 COVERAGE_INC(facet_unexpected);
4095 dpif_flow_del(ofproto->backer->dpif, key, key_len, NULL);
4098 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4100 * This function also pushes statistics updates to rules which each facet
4101 * resubmits into. Generally these statistics will be accurate. However, if a
4102 * facet changes the rule it resubmits into at some time in between
4103 * update_stats() runs, it is possible that statistics accrued to the
4104 * old rule will be incorrectly attributed to the new rule. This could be
4105 * avoided by calling update_stats() whenever rules are created or
4106 * deleted. However, the performance impact of making so many calls to the
4107 * datapath do not justify the benefit of having perfectly accurate statistics.
4110 update_stats(struct dpif_backer *backer)
4112 const struct dpif_flow_stats *stats;
4113 struct dpif_flow_dump dump;
4114 const struct nlattr *key;
4117 dpif_flow_dump_start(&dump, backer->dpif);
4118 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
4120 struct subfacet *subfacet;
4121 struct ofproto_dpif *ofproto;
4122 struct ofport_dpif *ofport;
4125 if (ofproto_receive(backer, NULL, key, key_len, &flow, NULL, &ofproto,
4130 ofport = get_ofp_port(ofproto, flow.in_port);
4131 if (ofport && ofport->tnl_port) {
4132 netdev_vport_inc_rx(ofport->up.netdev, stats);
4135 key_hash = odp_flow_key_hash(key, key_len);
4136 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
4137 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
4139 update_subfacet_stats(subfacet, stats);
4143 /* Stats are updated per-packet. */
4146 case SF_NOT_INSTALLED:
4148 delete_unexpected_flow(ofproto, key, key_len);
4152 dpif_flow_dump_done(&dump);
4155 /* Calculates and returns the number of milliseconds of idle time after which
4156 * subfacets should expire from the datapath. When a subfacet expires, we fold
4157 * its statistics into its facet, and when a facet's last subfacet expires, we
4158 * fold its statistic into its rule. */
4160 subfacet_max_idle(const struct ofproto_dpif *ofproto)
4163 * Idle time histogram.
4165 * Most of the time a switch has a relatively small number of subfacets.
4166 * When this is the case we might as well keep statistics for all of them
4167 * in userspace and to cache them in the kernel datapath for performance as
4170 * As the number of subfacets increases, the memory required to maintain
4171 * statistics about them in userspace and in the kernel becomes
4172 * significant. However, with a large number of subfacets it is likely
4173 * that only a few of them are "heavy hitters" that consume a large amount
4174 * of bandwidth. At this point, only heavy hitters are worth caching in
4175 * the kernel and maintaining in userspaces; other subfacets we can
4178 * The technique used to compute the idle time is to build a histogram with
4179 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
4180 * that is installed in the kernel gets dropped in the appropriate bucket.
4181 * After the histogram has been built, we compute the cutoff so that only
4182 * the most-recently-used 1% of subfacets (but at least
4183 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
4184 * the most-recently-used bucket of subfacets is kept, so actually an
4185 * arbitrary number of subfacets can be kept in any given expiration run
4186 * (though the next run will delete most of those unless they receive
4189 * This requires a second pass through the subfacets, in addition to the
4190 * pass made by update_stats(), because the former function never looks at
4191 * uninstallable subfacets.
4193 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4194 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4195 int buckets[N_BUCKETS] = { 0 };
4196 int total, subtotal, bucket;
4197 struct subfacet *subfacet;
4201 total = hmap_count(&ofproto->subfacets);
4202 if (total <= ofproto->up.flow_eviction_threshold) {
4203 return N_BUCKETS * BUCKET_WIDTH;
4206 /* Build histogram. */
4208 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
4209 long long int idle = now - subfacet->used;
4210 int bucket = (idle <= 0 ? 0
4211 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4212 : (unsigned int) idle / BUCKET_WIDTH);
4216 /* Find the first bucket whose flows should be expired. */
4217 subtotal = bucket = 0;
4219 subtotal += buckets[bucket++];
4220 } while (bucket < N_BUCKETS &&
4221 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
4223 if (VLOG_IS_DBG_ENABLED()) {
4227 ds_put_cstr(&s, "keep");
4228 for (i = 0; i < N_BUCKETS; i++) {
4230 ds_put_cstr(&s, ", drop");
4233 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4236 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
4240 return bucket * BUCKET_WIDTH;
4244 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
4246 /* Cutoff time for most flows. */
4247 long long int normal_cutoff = time_msec() - dp_max_idle;
4249 /* We really want to keep flows for special protocols around, so use a more
4250 * conservative cutoff. */
4251 long long int special_cutoff = time_msec() - 10000;
4253 struct subfacet *subfacet, *next_subfacet;
4254 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
4258 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
4259 &ofproto->subfacets) {
4260 long long int cutoff;
4262 cutoff = (subfacet->slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)
4265 if (subfacet->used < cutoff) {
4266 if (subfacet->path != SF_NOT_INSTALLED) {
4267 batch[n_batch++] = subfacet;
4268 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
4269 subfacet_destroy_batch(ofproto, batch, n_batch);
4273 subfacet_destroy(subfacet);
4279 subfacet_destroy_batch(ofproto, batch, n_batch);
4283 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4284 * then delete it entirely. */
4286 rule_expire(struct rule_dpif *rule)
4288 struct facet *facet, *next_facet;
4292 if (rule->up.pending) {
4293 /* We'll have to expire it later. */
4297 /* Has 'rule' expired? */
4299 if (rule->up.hard_timeout
4300 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
4301 reason = OFPRR_HARD_TIMEOUT;
4302 } else if (rule->up.idle_timeout
4303 && now > rule->up.used + rule->up.idle_timeout * 1000) {
4304 reason = OFPRR_IDLE_TIMEOUT;
4309 COVERAGE_INC(ofproto_dpif_expired);
4311 /* Update stats. (This is a no-op if the rule expired due to an idle
4312 * timeout, because that only happens when the rule has no facets left.) */
4313 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4314 facet_remove(facet);
4317 /* Get rid of the rule. */
4318 ofproto_rule_expire(&rule->up, reason);
4323 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
4325 * The caller must already have determined that no facet with an identical
4326 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
4327 * the ofproto's classifier table.
4329 * 'hash' must be the return value of flow_hash(flow, 0).
4331 * The facet will initially have no subfacets. The caller should create (at
4332 * least) one subfacet with subfacet_create(). */
4333 static struct facet *
4334 facet_create(struct rule_dpif *rule, const struct flow *flow, uint32_t hash)
4336 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4337 struct facet *facet;
4339 facet = xzalloc(sizeof *facet);
4340 facet->used = time_msec();
4341 hmap_insert(&ofproto->facets, &facet->hmap_node, hash);
4342 list_push_back(&rule->facets, &facet->list_node);
4344 facet->flow = *flow;
4345 list_init(&facet->subfacets);
4346 netflow_flow_init(&facet->nf_flow);
4347 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4353 facet_free(struct facet *facet)
4358 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4359 * 'packet', which arrived on 'in_port'. */
4361 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4362 const struct nlattr *odp_actions, size_t actions_len,
4363 struct ofpbuf *packet)
4365 struct odputil_keybuf keybuf;
4369 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4370 odp_flow_key_from_flow(&key, flow,
4371 ofp_port_to_odp_port(ofproto, flow->in_port));
4373 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4374 odp_actions, actions_len, packet);
4378 /* Remove 'facet' from 'ofproto' and free up the associated memory:
4380 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4381 * rule's statistics, via subfacet_uninstall().
4383 * - Removes 'facet' from its rule and from ofproto->facets.
4386 facet_remove(struct facet *facet)
4388 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4389 struct subfacet *subfacet, *next_subfacet;
4391 ovs_assert(!list_is_empty(&facet->subfacets));
4393 /* First uninstall all of the subfacets to get final statistics. */
4394 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4395 subfacet_uninstall(subfacet);
4398 /* Flush the final stats to the rule.
4400 * This might require us to have at least one subfacet around so that we
4401 * can use its actions for accounting in facet_account(), which is why we
4402 * have uninstalled but not yet destroyed the subfacets. */
4403 facet_flush_stats(facet);
4405 /* Now we're really all done so destroy everything. */
4406 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4407 &facet->subfacets) {
4408 subfacet_destroy__(subfacet);
4410 hmap_remove(&ofproto->facets, &facet->hmap_node);
4411 list_remove(&facet->list_node);
4415 /* Feed information from 'facet' back into the learning table to keep it in
4416 * sync with what is actually flowing through the datapath. */
4418 facet_learn(struct facet *facet)
4420 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4421 struct subfacet *subfacet= CONTAINER_OF(list_front(&facet->subfacets),
4422 struct subfacet, list_node);
4423 struct action_xlate_ctx ctx;
4425 if (!facet->has_learn
4426 && !facet->has_normal
4427 && (!facet->has_fin_timeout
4428 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4432 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4433 &subfacet->initial_vals,
4434 facet->rule, facet->tcp_flags, NULL);
4435 ctx.may_learn = true;
4436 xlate_actions_for_side_effects(&ctx, facet->rule->up.ofpacts,
4437 facet->rule->up.ofpacts_len);
4441 facet_account(struct facet *facet)
4443 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4444 struct subfacet *subfacet;
4445 const struct nlattr *a;
4450 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
4453 n_bytes = facet->byte_count - facet->accounted_bytes;
4455 /* This loop feeds byte counters to bond_account() for rebalancing to use
4456 * as a basis. We also need to track the actual VLAN on which the packet
4457 * is going to be sent to ensure that it matches the one passed to
4458 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4461 * We use the actions from an arbitrary subfacet because they should all
4462 * be equally valid for our purpose. */
4463 subfacet = CONTAINER_OF(list_front(&facet->subfacets),
4464 struct subfacet, list_node);
4465 vlan_tci = facet->flow.vlan_tci;
4466 NL_ATTR_FOR_EACH_UNSAFE (a, left,
4467 subfacet->actions, subfacet->actions_len) {
4468 const struct ovs_action_push_vlan *vlan;
4469 struct ofport_dpif *port;
4471 switch (nl_attr_type(a)) {
4472 case OVS_ACTION_ATTR_OUTPUT:
4473 port = get_odp_port(ofproto, nl_attr_get_u32(a));
4474 if (port && port->bundle && port->bundle->bond) {
4475 bond_account(port->bundle->bond, &facet->flow,
4476 vlan_tci_to_vid(vlan_tci), n_bytes);
4480 case OVS_ACTION_ATTR_POP_VLAN:
4481 vlan_tci = htons(0);
4484 case OVS_ACTION_ATTR_PUSH_VLAN:
4485 vlan = nl_attr_get(a);
4486 vlan_tci = vlan->vlan_tci;
4492 /* Returns true if the only action for 'facet' is to send to the controller.
4493 * (We don't report NetFlow expiration messages for such facets because they
4494 * are just part of the control logic for the network, not real traffic). */
4496 facet_is_controller_flow(struct facet *facet)
4499 const struct rule *rule = &facet->rule->up;
4500 const struct ofpact *ofpacts = rule->ofpacts;
4501 size_t ofpacts_len = rule->ofpacts_len;
4503 if (ofpacts_len > 0 &&
4504 ofpacts->type == OFPACT_CONTROLLER &&
4505 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4512 /* Folds all of 'facet''s statistics into its rule. Also updates the
4513 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4514 * 'facet''s statistics in the datapath should have been zeroed and folded into
4515 * its packet and byte counts before this function is called. */
4517 facet_flush_stats(struct facet *facet)
4519 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4520 struct subfacet *subfacet;
4522 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4523 ovs_assert(!subfacet->dp_byte_count);
4524 ovs_assert(!subfacet->dp_packet_count);
4527 facet_push_stats(facet);
4528 if (facet->accounted_bytes < facet->byte_count) {
4529 facet_account(facet);
4530 facet->accounted_bytes = facet->byte_count;
4533 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4534 struct ofexpired expired;
4535 expired.flow = facet->flow;
4536 expired.packet_count = facet->packet_count;
4537 expired.byte_count = facet->byte_count;
4538 expired.used = facet->used;
4539 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4542 facet->rule->packet_count += facet->packet_count;
4543 facet->rule->byte_count += facet->byte_count;
4545 /* Reset counters to prevent double counting if 'facet' ever gets
4547 facet_reset_counters(facet);
4549 netflow_flow_clear(&facet->nf_flow);
4550 facet->tcp_flags = 0;
4553 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4554 * Returns it if found, otherwise a null pointer.
4556 * 'hash' must be the return value of flow_hash(flow, 0).
4558 * The returned facet might need revalidation; use facet_lookup_valid()
4559 * instead if that is important. */
4560 static struct facet *
4561 facet_find(struct ofproto_dpif *ofproto,
4562 const struct flow *flow, uint32_t hash)
4564 struct facet *facet;
4566 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) {
4567 if (flow_equal(flow, &facet->flow)) {
4575 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4576 * Returns it if found, otherwise a null pointer.
4578 * 'hash' must be the return value of flow_hash(flow, 0).
4580 * The returned facet is guaranteed to be valid. */
4581 static struct facet *
4582 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow,
4585 struct facet *facet;
4587 facet = facet_find(ofproto, flow, hash);
4589 && (ofproto->backer->need_revalidate
4590 || tag_set_intersects(&ofproto->backer->revalidate_set,
4592 facet_revalidate(facet);
4594 /* facet_revalidate() may have destroyed 'facet'. */
4595 facet = facet_find(ofproto, flow, hash);
4602 subfacet_path_to_string(enum subfacet_path path)
4605 case SF_NOT_INSTALLED:
4606 return "not installed";
4608 return "in fast path";
4610 return "in slow path";
4616 /* Returns the path in which a subfacet should be installed if its 'slow'
4617 * member has the specified value. */
4618 static enum subfacet_path
4619 subfacet_want_path(enum slow_path_reason slow)
4621 return slow ? SF_SLOW_PATH : SF_FAST_PATH;
4624 /* Returns true if 'subfacet' needs to have its datapath flow updated,
4625 * supposing that its actions have been recalculated as 'want_actions' and that
4626 * 'slow' is nonzero iff 'subfacet' should be in the slow path. */
4628 subfacet_should_install(struct subfacet *subfacet, enum slow_path_reason slow,
4629 const struct ofpbuf *want_actions)
4631 enum subfacet_path want_path = subfacet_want_path(slow);
4632 return (want_path != subfacet->path
4633 || (want_path == SF_FAST_PATH
4634 && (subfacet->actions_len != want_actions->size
4635 || memcmp(subfacet->actions, want_actions->data,
4636 subfacet->actions_len))));
4640 facet_check_consistency(struct facet *facet)
4642 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4644 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4646 uint64_t odp_actions_stub[1024 / 8];
4647 struct ofpbuf odp_actions;
4649 struct rule_dpif *rule;
4650 struct subfacet *subfacet;
4651 bool may_log = false;
4654 /* Check the rule for consistency. */
4655 rule = rule_dpif_lookup(ofproto, &facet->flow);
4656 ok = rule == facet->rule;
4658 may_log = !VLOG_DROP_WARN(&rl);
4663 flow_format(&s, &facet->flow);
4664 ds_put_format(&s, ": facet associated with wrong rule (was "
4665 "table=%"PRIu8",", facet->rule->up.table_id);
4666 cls_rule_format(&facet->rule->up.cr, &s);
4667 ds_put_format(&s, ") (should have been table=%"PRIu8",",
4669 cls_rule_format(&rule->up.cr, &s);
4670 ds_put_char(&s, ')');
4672 VLOG_WARN("%s", ds_cstr(&s));
4677 /* Check the datapath actions for consistency. */
4678 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4679 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4680 enum subfacet_path want_path;
4681 struct action_xlate_ctx ctx;
4684 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4685 &subfacet->initial_vals, rule, 0, NULL);
4686 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
4689 if (subfacet->path == SF_NOT_INSTALLED) {
4690 /* This only happens if the datapath reported an error when we
4691 * tried to install the flow. Don't flag another error here. */
4695 want_path = subfacet_want_path(subfacet->slow);
4696 if (want_path == SF_SLOW_PATH && subfacet->path == SF_SLOW_PATH) {
4697 /* The actions for slow-path flows may legitimately vary from one
4698 * packet to the next. We're done. */
4702 if (!subfacet_should_install(subfacet, subfacet->slow, &odp_actions)) {
4706 /* Inconsistency! */
4708 may_log = !VLOG_DROP_WARN(&rl);
4712 /* Rate-limited, skip reporting. */
4717 odp_flow_key_format(subfacet->key, subfacet->key_len, &s);
4719 ds_put_cstr(&s, ": inconsistency in subfacet");
4720 if (want_path != subfacet->path) {
4721 enum odp_key_fitness fitness = subfacet->key_fitness;
4723 ds_put_format(&s, " (%s, fitness=%s)",
4724 subfacet_path_to_string(subfacet->path),
4725 odp_key_fitness_to_string(fitness));
4726 ds_put_format(&s, " (should have been %s)",
4727 subfacet_path_to_string(want_path));
4728 } else if (want_path == SF_FAST_PATH) {
4729 ds_put_cstr(&s, " (actions were: ");
4730 format_odp_actions(&s, subfacet->actions,
4731 subfacet->actions_len);
4732 ds_put_cstr(&s, ") (correct actions: ");
4733 format_odp_actions(&s, odp_actions.data, odp_actions.size);
4734 ds_put_char(&s, ')');
4736 ds_put_cstr(&s, " (actions: ");
4737 format_odp_actions(&s, subfacet->actions,
4738 subfacet->actions_len);
4739 ds_put_char(&s, ')');
4741 VLOG_WARN("%s", ds_cstr(&s));
4744 ofpbuf_uninit(&odp_actions);
4749 /* Re-searches the classifier for 'facet':
4751 * - If the rule found is different from 'facet''s current rule, moves
4752 * 'facet' to the new rule and recompiles its actions.
4754 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4755 * where it is and recompiles its actions anyway.
4757 * - If any of 'facet''s subfacets correspond to a new flow according to
4758 * ofproto_receive(), 'facet' is removed. */
4760 facet_revalidate(struct facet *facet)
4762 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4764 struct nlattr *odp_actions;
4767 struct actions *new_actions;
4769 struct action_xlate_ctx ctx;
4770 uint64_t odp_actions_stub[1024 / 8];
4771 struct ofpbuf odp_actions;
4773 struct rule_dpif *new_rule;
4774 struct subfacet *subfacet;
4777 COVERAGE_INC(facet_revalidate);
4779 /* Check that child subfacets still correspond to this facet. Tunnel
4780 * configuration changes could cause a subfacet's OpenFlow in_port to
4782 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4783 struct ofproto_dpif *recv_ofproto;
4784 struct flow recv_flow;
4787 error = ofproto_receive(ofproto->backer, NULL, subfacet->key,
4788 subfacet->key_len, &recv_flow, NULL,
4789 &recv_ofproto, NULL, NULL);
4791 || recv_ofproto != ofproto
4792 || memcmp(&recv_flow, &facet->flow, sizeof recv_flow)) {
4793 facet_remove(facet);
4798 new_rule = rule_dpif_lookup(ofproto, &facet->flow);
4800 /* Calculate new datapath actions.
4802 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4803 * emit a NetFlow expiration and, if so, we need to have the old state
4804 * around to properly compose it. */
4806 /* If the datapath actions changed or the installability changed,
4807 * then we need to talk to the datapath. */
4810 memset(&ctx, 0, sizeof ctx);
4811 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4812 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4813 enum slow_path_reason slow;
4815 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4816 &subfacet->initial_vals, new_rule, 0, NULL);
4817 xlate_actions(&ctx, new_rule->up.ofpacts, new_rule->up.ofpacts_len,
4820 slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4821 if (subfacet_should_install(subfacet, slow, &odp_actions)) {
4822 struct dpif_flow_stats stats;
4824 subfacet_install(subfacet,
4825 odp_actions.data, odp_actions.size, &stats, slow);
4826 subfacet_update_stats(subfacet, &stats);
4829 new_actions = xcalloc(list_size(&facet->subfacets),
4830 sizeof *new_actions);
4832 new_actions[i].odp_actions = xmemdup(odp_actions.data,
4834 new_actions[i].actions_len = odp_actions.size;
4839 ofpbuf_uninit(&odp_actions);
4842 facet_flush_stats(facet);
4845 /* Update 'facet' now that we've taken care of all the old state. */
4846 facet->tags = ctx.tags;
4847 facet->nf_flow.output_iface = ctx.nf_output_iface;
4848 facet->has_learn = ctx.has_learn;
4849 facet->has_normal = ctx.has_normal;
4850 facet->has_fin_timeout = ctx.has_fin_timeout;
4851 facet->mirrors = ctx.mirrors;
4854 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4855 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4857 if (new_actions && new_actions[i].odp_actions) {
4858 free(subfacet->actions);
4859 subfacet->actions = new_actions[i].odp_actions;
4860 subfacet->actions_len = new_actions[i].actions_len;
4866 if (facet->rule != new_rule) {
4867 COVERAGE_INC(facet_changed_rule);
4868 list_remove(&facet->list_node);
4869 list_push_back(&new_rule->facets, &facet->list_node);
4870 facet->rule = new_rule;
4871 facet->used = new_rule->up.created;
4872 facet->prev_used = facet->used;
4876 /* Updates 'facet''s used time. Caller is responsible for calling
4877 * facet_push_stats() to update the flows which 'facet' resubmits into. */
4879 facet_update_time(struct facet *facet, long long int used)
4881 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4882 if (used > facet->used) {
4884 ofproto_rule_update_used(&facet->rule->up, used);
4885 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
4890 facet_reset_counters(struct facet *facet)
4892 facet->packet_count = 0;
4893 facet->byte_count = 0;
4894 facet->prev_packet_count = 0;
4895 facet->prev_byte_count = 0;
4896 facet->accounted_bytes = 0;
4900 facet_push_stats(struct facet *facet)
4902 struct dpif_flow_stats stats;
4904 ovs_assert(facet->packet_count >= facet->prev_packet_count);
4905 ovs_assert(facet->byte_count >= facet->prev_byte_count);
4906 ovs_assert(facet->used >= facet->prev_used);
4908 stats.n_packets = facet->packet_count - facet->prev_packet_count;
4909 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4910 stats.used = facet->used;
4911 stats.tcp_flags = 0;
4913 if (stats.n_packets || stats.n_bytes || facet->used > facet->prev_used) {
4914 facet->prev_packet_count = facet->packet_count;
4915 facet->prev_byte_count = facet->byte_count;
4916 facet->prev_used = facet->used;
4918 flow_push_stats(facet, &stats);
4920 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
4921 facet->mirrors, stats.n_packets, stats.n_bytes);
4926 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
4928 rule->packet_count += stats->n_packets;
4929 rule->byte_count += stats->n_bytes;
4930 ofproto_rule_update_used(&rule->up, stats->used);
4933 /* Pushes flow statistics to the rules which 'facet->flow' resubmits
4934 * into given 'facet->rule''s actions and mirrors. */
4936 flow_push_stats(struct facet *facet, const struct dpif_flow_stats *stats)
4938 struct rule_dpif *rule = facet->rule;
4939 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4940 struct subfacet *subfacet = CONTAINER_OF(list_front(&facet->subfacets),
4941 struct subfacet, list_node);
4942 struct action_xlate_ctx ctx;
4944 ofproto_rule_update_used(&rule->up, stats->used);
4946 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4947 &subfacet->initial_vals, rule, 0, NULL);
4948 ctx.resubmit_stats = stats;
4949 xlate_actions_for_side_effects(&ctx, rule->up.ofpacts,
4950 rule->up.ofpacts_len);
4955 static struct subfacet *
4956 subfacet_find(struct ofproto_dpif *ofproto,
4957 const struct nlattr *key, size_t key_len, uint32_t key_hash)
4959 struct subfacet *subfacet;
4961 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4962 &ofproto->subfacets) {
4963 if (subfacet->key_len == key_len
4964 && !memcmp(key, subfacet->key, key_len)) {
4972 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4973 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
4974 * existing subfacet if there is one, otherwise creates and returns a
4977 * If the returned subfacet is new, then subfacet->actions will be NULL, in
4978 * which case the caller must populate the actions with
4979 * subfacet_make_actions(). */
4980 static struct subfacet *
4981 subfacet_create(struct facet *facet, struct flow_miss *miss,
4984 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4985 enum odp_key_fitness key_fitness = miss->key_fitness;
4986 const struct nlattr *key = miss->key;
4987 size_t key_len = miss->key_len;
4989 struct subfacet *subfacet;
4991 key_hash = odp_flow_key_hash(key, key_len);
4993 if (list_is_empty(&facet->subfacets)) {
4994 subfacet = &facet->one_subfacet;
4996 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
4998 if (subfacet->facet == facet) {
5002 /* This shouldn't happen. */
5003 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
5004 subfacet_destroy(subfacet);
5007 subfacet = xmalloc(sizeof *subfacet);
5010 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
5011 list_push_back(&facet->subfacets, &subfacet->list_node);
5012 subfacet->facet = facet;
5013 subfacet->key_fitness = key_fitness;
5014 subfacet->key = xmemdup(key, key_len);
5015 subfacet->key_len = key_len;
5016 subfacet->used = now;
5017 subfacet->dp_packet_count = 0;
5018 subfacet->dp_byte_count = 0;
5019 subfacet->actions_len = 0;
5020 subfacet->actions = NULL;
5021 subfacet->slow = (subfacet->key_fitness == ODP_FIT_TOO_LITTLE
5024 subfacet->path = SF_NOT_INSTALLED;
5025 subfacet->initial_vals = miss->initial_vals;
5026 subfacet->odp_in_port = miss->odp_in_port;
5031 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
5032 * its facet within 'ofproto', and frees it. */
5034 subfacet_destroy__(struct subfacet *subfacet)
5036 struct facet *facet = subfacet->facet;
5037 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5039 subfacet_uninstall(subfacet);
5040 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
5041 list_remove(&subfacet->list_node);
5042 free(subfacet->key);
5043 free(subfacet->actions);
5044 if (subfacet != &facet->one_subfacet) {
5049 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
5050 * last remaining subfacet in its facet destroys the facet too. */
5052 subfacet_destroy(struct subfacet *subfacet)
5054 struct facet *facet = subfacet->facet;
5056 if (list_is_singleton(&facet->subfacets)) {
5057 /* facet_remove() needs at least one subfacet (it will remove it). */
5058 facet_remove(facet);
5060 subfacet_destroy__(subfacet);
5065 subfacet_destroy_batch(struct ofproto_dpif *ofproto,
5066 struct subfacet **subfacets, int n)
5068 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
5069 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
5070 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
5073 for (i = 0; i < n; i++) {
5074 ops[i].type = DPIF_OP_FLOW_DEL;
5075 ops[i].u.flow_del.key = subfacets[i]->key;
5076 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
5077 ops[i].u.flow_del.stats = &stats[i];
5081 dpif_operate(ofproto->backer->dpif, opsp, n);
5082 for (i = 0; i < n; i++) {
5083 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
5084 subfacets[i]->path = SF_NOT_INSTALLED;
5085 subfacet_destroy(subfacets[i]);
5089 /* Composes the datapath actions for 'subfacet' based on its rule's actions.
5090 * Translates the actions into 'odp_actions', which the caller must have
5091 * initialized and is responsible for uninitializing. */
5093 subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet,
5094 struct ofpbuf *odp_actions)
5096 struct facet *facet = subfacet->facet;
5097 struct rule_dpif *rule = facet->rule;
5098 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5100 struct action_xlate_ctx ctx;
5102 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
5103 &subfacet->initial_vals, rule, 0, packet);
5104 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, odp_actions);
5105 facet->tags = ctx.tags;
5106 facet->has_learn = ctx.has_learn;
5107 facet->has_normal = ctx.has_normal;
5108 facet->has_fin_timeout = ctx.has_fin_timeout;
5109 facet->nf_flow.output_iface = ctx.nf_output_iface;
5110 facet->mirrors = ctx.mirrors;
5112 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
5113 if (subfacet->actions_len != odp_actions->size
5114 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
5115 free(subfacet->actions);
5116 subfacet->actions_len = odp_actions->size;
5117 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
5121 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
5122 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
5123 * in the datapath will be zeroed and 'stats' will be updated with traffic new
5124 * since 'subfacet' was last updated.
5126 * Returns 0 if successful, otherwise a positive errno value. */
5128 subfacet_install(struct subfacet *subfacet,
5129 const struct nlattr *actions, size_t actions_len,
5130 struct dpif_flow_stats *stats,
5131 enum slow_path_reason slow)
5133 struct facet *facet = subfacet->facet;
5134 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5135 enum subfacet_path path = subfacet_want_path(slow);
5136 uint64_t slow_path_stub[128 / 8];
5137 enum dpif_flow_put_flags flags;
5140 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
5142 flags |= DPIF_FP_ZERO_STATS;
5145 if (path == SF_SLOW_PATH) {
5146 compose_slow_path(ofproto, &facet->flow, slow,
5147 slow_path_stub, sizeof slow_path_stub,
5148 &actions, &actions_len);
5151 ret = dpif_flow_put(ofproto->backer->dpif, flags, subfacet->key,
5152 subfacet->key_len, actions, actions_len, stats);
5155 subfacet_reset_dp_stats(subfacet, stats);
5159 subfacet->path = path;
5165 subfacet_reinstall(struct subfacet *subfacet, struct dpif_flow_stats *stats)
5167 return subfacet_install(subfacet, subfacet->actions, subfacet->actions_len,
5168 stats, subfacet->slow);
5171 /* If 'subfacet' is installed in the datapath, uninstalls it. */
5173 subfacet_uninstall(struct subfacet *subfacet)
5175 if (subfacet->path != SF_NOT_INSTALLED) {
5176 struct rule_dpif *rule = subfacet->facet->rule;
5177 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5178 struct dpif_flow_stats stats;
5181 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
5182 subfacet->key_len, &stats);
5183 subfacet_reset_dp_stats(subfacet, &stats);
5185 subfacet_update_stats(subfacet, &stats);
5187 subfacet->path = SF_NOT_INSTALLED;
5189 ovs_assert(subfacet->dp_packet_count == 0);
5190 ovs_assert(subfacet->dp_byte_count == 0);
5194 /* Resets 'subfacet''s datapath statistics counters. This should be called
5195 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
5196 * non-null, it should contain the statistics returned by dpif when 'subfacet'
5197 * was reset in the datapath. 'stats' will be modified to include only
5198 * statistics new since 'subfacet' was last updated. */
5200 subfacet_reset_dp_stats(struct subfacet *subfacet,
5201 struct dpif_flow_stats *stats)
5204 && subfacet->dp_packet_count <= stats->n_packets
5205 && subfacet->dp_byte_count <= stats->n_bytes) {
5206 stats->n_packets -= subfacet->dp_packet_count;
5207 stats->n_bytes -= subfacet->dp_byte_count;
5210 subfacet->dp_packet_count = 0;
5211 subfacet->dp_byte_count = 0;
5214 /* Updates 'subfacet''s used time. The caller is responsible for calling
5215 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
5217 subfacet_update_time(struct subfacet *subfacet, long long int used)
5219 if (used > subfacet->used) {
5220 subfacet->used = used;
5221 facet_update_time(subfacet->facet, used);
5225 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
5227 * Because of the meaning of a subfacet's counters, it only makes sense to do
5228 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
5229 * represents a packet that was sent by hand or if it represents statistics
5230 * that have been cleared out of the datapath. */
5232 subfacet_update_stats(struct subfacet *subfacet,
5233 const struct dpif_flow_stats *stats)
5235 if (stats->n_packets || stats->used > subfacet->used) {
5236 struct facet *facet = subfacet->facet;
5238 subfacet_update_time(subfacet, stats->used);
5239 facet->packet_count += stats->n_packets;
5240 facet->byte_count += stats->n_bytes;
5241 facet->tcp_flags |= stats->tcp_flags;
5242 facet_push_stats(facet);
5243 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
5249 static struct rule_dpif *
5250 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow)
5252 struct rule_dpif *rule;
5254 rule = rule_dpif_lookup__(ofproto, flow, 0);
5259 return rule_dpif_miss_rule(ofproto, flow);
5262 static struct rule_dpif *
5263 rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow,
5266 struct cls_rule *cls_rule;
5267 struct classifier *cls;
5269 if (table_id >= N_TABLES) {
5273 cls = &ofproto->up.tables[table_id].cls;
5274 if (flow->nw_frag & FLOW_NW_FRAG_ANY
5275 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
5276 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
5277 * are unavailable. */
5278 struct flow ofpc_normal_flow = *flow;
5279 ofpc_normal_flow.tp_src = htons(0);
5280 ofpc_normal_flow.tp_dst = htons(0);
5281 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
5283 cls_rule = classifier_lookup(cls, flow);
5285 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
5288 static struct rule_dpif *
5289 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
5291 struct ofport_dpif *port;
5293 port = get_ofp_port(ofproto, flow->in_port);
5295 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port);
5296 return ofproto->miss_rule;
5299 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
5300 return ofproto->no_packet_in_rule;
5302 return ofproto->miss_rule;
5306 complete_operation(struct rule_dpif *rule)
5308 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5310 rule_invalidate(rule);
5312 struct dpif_completion *c = xmalloc(sizeof *c);
5313 c->op = rule->up.pending;
5314 list_push_back(&ofproto->completions, &c->list_node);
5316 ofoperation_complete(rule->up.pending, 0);
5320 static struct rule *
5323 struct rule_dpif *rule = xmalloc(sizeof *rule);
5328 rule_dealloc(struct rule *rule_)
5330 struct rule_dpif *rule = rule_dpif_cast(rule_);
5335 rule_construct(struct rule *rule_)
5337 struct rule_dpif *rule = rule_dpif_cast(rule_);
5338 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5339 struct rule_dpif *victim;
5342 rule->packet_count = 0;
5343 rule->byte_count = 0;
5345 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
5346 if (victim && !list_is_empty(&victim->facets)) {
5347 struct facet *facet;
5349 rule->facets = victim->facets;
5350 list_moved(&rule->facets);
5351 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5352 /* XXX: We're only clearing our local counters here. It's possible
5353 * that quite a few packets are unaccounted for in the datapath
5354 * statistics. These will be accounted to the new rule instead of
5355 * cleared as required. This could be fixed by clearing out the
5356 * datapath statistics for this facet, but currently it doesn't
5358 facet_reset_counters(facet);
5362 /* Must avoid list_moved() in this case. */
5363 list_init(&rule->facets);
5366 table_id = rule->up.table_id;
5368 rule->tag = victim->tag;
5369 } else if (table_id == 0) {
5374 miniflow_expand(&rule->up.cr.match.flow, &flow);
5375 rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask,
5376 ofproto->tables[table_id].basis);
5379 complete_operation(rule);
5384 rule_destruct(struct rule *rule_)
5386 struct rule_dpif *rule = rule_dpif_cast(rule_);
5387 struct facet *facet, *next_facet;
5389 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
5390 facet_revalidate(facet);
5393 complete_operation(rule);
5397 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5399 struct rule_dpif *rule = rule_dpif_cast(rule_);
5400 struct facet *facet;
5402 /* Start from historical data for 'rule' itself that are no longer tracked
5403 * in facets. This counts, for example, facets that have expired. */
5404 *packets = rule->packet_count;
5405 *bytes = rule->byte_count;
5407 /* Add any statistics that are tracked by facets. This includes
5408 * statistical data recently updated by ofproto_update_stats() as well as
5409 * stats for packets that were executed "by hand" via dpif_execute(). */
5410 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5411 *packets += facet->packet_count;
5412 *bytes += facet->byte_count;
5417 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
5418 struct ofpbuf *packet)
5420 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5421 struct initial_vals initial_vals;
5422 struct dpif_flow_stats stats;
5423 struct action_xlate_ctx ctx;
5424 uint64_t odp_actions_stub[1024 / 8];
5425 struct ofpbuf odp_actions;
5427 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5428 rule_credit_stats(rule, &stats);
5430 initial_vals.vlan_tci = flow->vlan_tci;
5431 initial_vals.tunnel_ip_tos = flow->tunnel.ip_tos;
5432 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5433 action_xlate_ctx_init(&ctx, ofproto, flow, &initial_vals,
5434 rule, stats.tcp_flags, packet);
5435 ctx.resubmit_stats = &stats;
5436 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, &odp_actions);
5438 execute_odp_actions(ofproto, flow, odp_actions.data,
5439 odp_actions.size, packet);
5441 ofpbuf_uninit(&odp_actions);
5445 rule_execute(struct rule *rule, const struct flow *flow,
5446 struct ofpbuf *packet)
5448 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
5449 ofpbuf_delete(packet);
5454 rule_modify_actions(struct rule *rule_)
5456 struct rule_dpif *rule = rule_dpif_cast(rule_);
5458 complete_operation(rule);
5461 /* Sends 'packet' out 'ofport'.
5462 * May modify 'packet'.
5463 * Returns 0 if successful, otherwise a positive errno value. */
5465 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5467 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5468 uint64_t odp_actions_stub[1024 / 8];
5469 struct ofpbuf key, odp_actions;
5470 struct odputil_keybuf keybuf;
5475 flow_extract(packet, 0, 0, NULL, OFPP_LOCAL, &flow);
5476 if (netdev_vport_is_patch(ofport->up.netdev)) {
5477 struct ofproto_dpif *peer_ofproto;
5478 struct dpif_flow_stats stats;
5479 struct ofport_dpif *peer;
5480 struct rule_dpif *rule;
5482 peer = ofport_get_peer(ofport);
5487 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5488 netdev_vport_inc_tx(ofport->up.netdev, &stats);
5489 netdev_vport_inc_rx(peer->up.netdev, &stats);
5491 flow.in_port = peer->up.ofp_port;
5492 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
5493 rule = rule_dpif_lookup(peer_ofproto, &flow);
5494 rule_dpif_execute(rule, &flow, packet);
5499 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5501 if (ofport->tnl_port) {
5502 struct dpif_flow_stats stats;
5504 odp_port = tnl_port_send(ofport->tnl_port, &flow);
5505 if (odp_port == OVSP_NONE) {
5509 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5510 netdev_vport_inc_tx(ofport->up.netdev, &stats);
5511 odp_put_tunnel_action(&flow.tunnel, &odp_actions);
5512 odp_put_skb_mark_action(flow.skb_mark, &odp_actions);
5514 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
5516 if (odp_port != ofport->odp_port) {
5517 eth_pop_vlan(packet);
5518 flow.vlan_tci = htons(0);
5522 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5523 odp_flow_key_from_flow(&key, &flow,
5524 ofp_port_to_odp_port(ofproto, flow.in_port));
5526 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
5528 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
5529 error = dpif_execute(ofproto->backer->dpif,
5531 odp_actions.data, odp_actions.size,
5533 ofpbuf_uninit(&odp_actions);
5536 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
5537 ofproto->up.name, odp_port, strerror(error));
5539 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
5543 /* OpenFlow to datapath action translation. */
5545 static bool may_receive(const struct ofport_dpif *, struct action_xlate_ctx *);
5546 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
5547 struct action_xlate_ctx *);
5548 static void xlate_normal(struct action_xlate_ctx *);
5550 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5551 * The action will state 'slow' as the reason that the action is in the slow
5552 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5553 * dump-flows" output to see why a flow is in the slow path.)
5555 * The 'stub_size' bytes in 'stub' will be used to store the action.
5556 * 'stub_size' must be large enough for the action.
5558 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5561 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5562 enum slow_path_reason slow,
5563 uint64_t *stub, size_t stub_size,
5564 const struct nlattr **actionsp, size_t *actions_lenp)
5566 union user_action_cookie cookie;
5569 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5570 cookie.slow_path.unused = 0;
5571 cookie.slow_path.reason = slow;
5573 ofpbuf_use_stack(&buf, stub, stub_size);
5574 if (slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)) {
5575 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif, UINT32_MAX);
5576 odp_put_userspace_action(pid, &cookie, sizeof cookie, &buf);
5578 put_userspace_action(ofproto, &buf, flow, &cookie);
5580 *actionsp = buf.data;
5581 *actions_lenp = buf.size;
5585 put_userspace_action(const struct ofproto_dpif *ofproto,
5586 struct ofpbuf *odp_actions,
5587 const struct flow *flow,
5588 const union user_action_cookie *cookie)
5592 pid = dpif_port_get_pid(ofproto->backer->dpif,
5593 ofp_port_to_odp_port(ofproto, flow->in_port));
5595 return odp_put_userspace_action(pid, cookie, sizeof *cookie, odp_actions);
5599 compose_sflow_cookie(const struct ofproto_dpif *ofproto,
5600 ovs_be16 vlan_tci, uint32_t odp_port,
5601 unsigned int n_outputs, union user_action_cookie *cookie)
5605 cookie->type = USER_ACTION_COOKIE_SFLOW;
5606 cookie->sflow.vlan_tci = vlan_tci;
5608 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
5609 * port information") for the interpretation of cookie->output. */
5610 switch (n_outputs) {
5612 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
5613 cookie->sflow.output = 0x40000000 | 256;
5617 ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
5619 cookie->sflow.output = ifindex;
5624 /* 0x80000000 means "multiple output ports. */
5625 cookie->sflow.output = 0x80000000 | n_outputs;
5630 /* Compose SAMPLE action for sFlow. */
5632 compose_sflow_action(const struct ofproto_dpif *ofproto,
5633 struct ofpbuf *odp_actions,
5634 const struct flow *flow,
5637 uint32_t probability;
5638 union user_action_cookie cookie;
5639 size_t sample_offset, actions_offset;
5642 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
5646 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
5648 /* Number of packets out of UINT_MAX to sample. */
5649 probability = dpif_sflow_get_probability(ofproto->sflow);
5650 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
5652 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
5653 compose_sflow_cookie(ofproto, htons(0), odp_port,
5654 odp_port == OVSP_NONE ? 0 : 1, &cookie);
5655 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
5657 nl_msg_end_nested(odp_actions, actions_offset);
5658 nl_msg_end_nested(odp_actions, sample_offset);
5659 return cookie_offset;
5662 /* SAMPLE action must be first action in any given list of actions.
5663 * At this point we do not have all information required to build it. So try to
5664 * build sample action as complete as possible. */
5666 add_sflow_action(struct action_xlate_ctx *ctx)
5668 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
5670 &ctx->flow, OVSP_NONE);
5671 ctx->sflow_odp_port = 0;
5672 ctx->sflow_n_outputs = 0;
5675 /* Fix SAMPLE action according to data collected while composing ODP actions.
5676 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
5677 * USERSPACE action's user-cookie which is required for sflow. */
5679 fix_sflow_action(struct action_xlate_ctx *ctx)
5681 const struct flow *base = &ctx->base_flow;
5682 union user_action_cookie *cookie;
5684 if (!ctx->user_cookie_offset) {
5688 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
5690 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
5692 compose_sflow_cookie(ctx->ofproto, base->vlan_tci,
5693 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
5697 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
5700 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
5701 ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
5702 ovs_be64 flow_tun_id = ctx->flow.tunnel.tun_id;
5703 uint8_t flow_nw_tos = ctx->flow.nw_tos;
5704 struct priority_to_dscp *pdscp;
5705 uint32_t out_port, odp_port;
5707 /* If 'struct flow' gets additional metadata, we'll need to zero it out
5708 * before traversing a patch port. */
5709 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 19);
5712 xlate_report(ctx, "Nonexistent output port");
5714 } else if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD) {
5715 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
5717 } else if (check_stp && !stp_forward_in_state(ofport->stp_state)) {
5718 xlate_report(ctx, "STP not in forwarding state, skipping output");
5722 if (netdev_vport_is_patch(ofport->up.netdev)) {
5723 struct ofport_dpif *peer = ofport_get_peer(ofport);
5724 struct flow old_flow = ctx->flow;
5725 const struct ofproto_dpif *peer_ofproto;
5726 enum slow_path_reason special;
5727 struct ofport_dpif *in_port;
5730 xlate_report(ctx, "Nonexistent patch port peer");
5734 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
5735 if (peer_ofproto->backer != ctx->ofproto->backer) {
5736 xlate_report(ctx, "Patch port peer on a different datapath");
5740 ctx->ofproto = ofproto_dpif_cast(peer->up.ofproto);
5741 ctx->flow.in_port = peer->up.ofp_port;
5742 ctx->flow.metadata = htonll(0);
5743 memset(&ctx->flow.tunnel, 0, sizeof ctx->flow.tunnel);
5744 memset(ctx->flow.regs, 0, sizeof ctx->flow.regs);
5746 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
5747 special = process_special(ctx->ofproto, &ctx->flow, in_port,
5750 ctx->slow |= special;
5751 } else if (!in_port || may_receive(in_port, ctx)) {
5752 if (!in_port || stp_forward_in_state(in_port->stp_state)) {
5753 xlate_table_action(ctx, ctx->flow.in_port, 0, true);
5755 /* Forwarding is disabled by STP. Let OFPP_NORMAL and the
5756 * learning action look at the packet, then drop it. */
5757 struct flow old_base_flow = ctx->base_flow;
5758 size_t old_size = ctx->odp_actions->size;
5759 xlate_table_action(ctx, ctx->flow.in_port, 0, true);
5760 ctx->base_flow = old_base_flow;
5761 ctx->odp_actions->size = old_size;
5765 ctx->flow = old_flow;
5766 ctx->ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5768 if (ctx->resubmit_stats) {
5769 netdev_vport_inc_tx(ofport->up.netdev, ctx->resubmit_stats);
5770 netdev_vport_inc_rx(peer->up.netdev, ctx->resubmit_stats);
5776 pdscp = get_priority(ofport, ctx->flow.skb_priority);
5778 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
5779 ctx->flow.nw_tos |= pdscp->dscp;
5782 odp_port = ofp_port_to_odp_port(ctx->ofproto, ofp_port);
5783 if (ofport->tnl_port) {
5784 odp_port = tnl_port_send(ofport->tnl_port, &ctx->flow);
5785 if (odp_port == OVSP_NONE) {
5786 xlate_report(ctx, "Tunneling decided against output");
5790 if (ctx->resubmit_stats) {
5791 netdev_vport_inc_tx(ofport->up.netdev, ctx->resubmit_stats);
5793 out_port = odp_port;
5794 commit_odp_tunnel_action(&ctx->flow, &ctx->base_flow,
5797 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
5798 ctx->flow.vlan_tci);
5799 if (out_port != odp_port) {
5800 ctx->flow.vlan_tci = htons(0);
5803 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
5804 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
5806 ctx->sflow_odp_port = odp_port;
5807 ctx->sflow_n_outputs++;
5808 ctx->nf_output_iface = ofp_port;
5809 ctx->flow.tunnel.tun_id = flow_tun_id;
5810 ctx->flow.vlan_tci = flow_vlan_tci;
5811 ctx->flow.nw_tos = flow_nw_tos;
5815 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
5817 compose_output_action__(ctx, ofp_port, true);
5821 xlate_table_action(struct action_xlate_ctx *ctx,
5822 uint16_t in_port, uint8_t table_id, bool may_packet_in)
5824 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
5825 struct ofproto_dpif *ofproto = ctx->ofproto;
5826 struct rule_dpif *rule;
5827 uint16_t old_in_port;
5828 uint8_t old_table_id;
5830 old_table_id = ctx->table_id;
5831 ctx->table_id = table_id;
5833 /* Look up a flow with 'in_port' as the input port. */
5834 old_in_port = ctx->flow.in_port;
5835 ctx->flow.in_port = in_port;
5836 rule = rule_dpif_lookup__(ofproto, &ctx->flow, table_id);
5839 if (table_id > 0 && table_id < N_TABLES) {
5840 struct table_dpif *table = &ofproto->tables[table_id];
5841 if (table->other_table) {
5842 ctx->tags |= (rule && rule->tag
5844 : rule_calculate_tag(&ctx->flow,
5845 &table->other_table->mask,
5850 /* Restore the original input port. Otherwise OFPP_NORMAL and
5851 * OFPP_IN_PORT will have surprising behavior. */
5852 ctx->flow.in_port = old_in_port;
5854 if (ctx->resubmit_hook) {
5855 ctx->resubmit_hook(ctx, rule);
5858 if (rule == NULL && may_packet_in) {
5860 * check if table configuration flags
5861 * OFPTC_TABLE_MISS_CONTROLLER, default.
5862 * OFPTC_TABLE_MISS_CONTINUE,
5863 * OFPTC_TABLE_MISS_DROP
5864 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do?
5866 rule = rule_dpif_miss_rule(ofproto, &ctx->flow);
5870 struct rule_dpif *old_rule = ctx->rule;
5872 if (ctx->resubmit_stats) {
5873 rule_credit_stats(rule, ctx->resubmit_stats);
5878 do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx);
5879 ctx->rule = old_rule;
5883 ctx->table_id = old_table_id;
5885 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
5887 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
5888 MAX_RESUBMIT_RECURSION);
5889 ctx->max_resubmit_trigger = true;
5894 xlate_ofpact_resubmit(struct action_xlate_ctx *ctx,
5895 const struct ofpact_resubmit *resubmit)
5900 in_port = resubmit->in_port;
5901 if (in_port == OFPP_IN_PORT) {
5902 in_port = ctx->flow.in_port;
5905 table_id = resubmit->table_id;
5906 if (table_id == 255) {
5907 table_id = ctx->table_id;
5910 xlate_table_action(ctx, in_port, table_id, false);
5914 flood_packets(struct action_xlate_ctx *ctx, bool all)
5916 struct ofport_dpif *ofport;
5918 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
5919 uint16_t ofp_port = ofport->up.ofp_port;
5921 if (ofp_port == ctx->flow.in_port) {
5926 compose_output_action__(ctx, ofp_port, false);
5927 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
5928 compose_output_action(ctx, ofp_port);
5932 ctx->nf_output_iface = NF_OUT_FLOOD;
5936 execute_controller_action(struct action_xlate_ctx *ctx, int len,
5937 enum ofp_packet_in_reason reason,
5938 uint16_t controller_id)
5940 struct ofputil_packet_in pin;
5941 struct ofpbuf *packet;
5943 ctx->slow |= SLOW_CONTROLLER;
5948 packet = ofpbuf_clone(ctx->packet);
5950 if (packet->l2 && packet->l3) {
5951 struct eth_header *eh;
5952 uint16_t mpls_depth;
5954 eth_pop_vlan(packet);
5957 memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
5958 memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
5960 if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
5961 eth_push_vlan(packet, ctx->flow.vlan_tci);
5964 mpls_depth = eth_mpls_depth(packet);
5966 if (mpls_depth < ctx->flow.mpls_depth) {
5967 push_mpls(packet, ctx->flow.dl_type, ctx->flow.mpls_lse);
5968 } else if (mpls_depth > ctx->flow.mpls_depth) {
5969 pop_mpls(packet, ctx->flow.dl_type);
5970 } else if (mpls_depth) {
5971 set_mpls_lse(packet, ctx->flow.mpls_lse);
5975 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
5976 packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
5977 ctx->flow.nw_tos, ctx->flow.nw_ttl);
5981 if (ctx->flow.nw_proto == IPPROTO_TCP) {
5982 packet_set_tcp_port(packet, ctx->flow.tp_src,
5984 } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
5985 packet_set_udp_port(packet, ctx->flow.tp_src,
5992 pin.packet = packet->data;
5993 pin.packet_len = packet->size;
5994 pin.reason = reason;
5995 pin.controller_id = controller_id;
5996 pin.table_id = ctx->table_id;
5997 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
6000 flow_get_metadata(&ctx->flow, &pin.fmd);
6002 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
6003 ofpbuf_delete(packet);
6007 execute_mpls_push_action(struct action_xlate_ctx *ctx, ovs_be16 eth_type)
6009 ovs_assert(eth_type_mpls(eth_type));
6011 if (ctx->base_flow.mpls_depth) {
6012 ctx->flow.mpls_lse &= ~htonl(MPLS_BOS_MASK);
6013 ctx->flow.mpls_depth++;
6018 if (ctx->flow.dl_type == htons(ETH_TYPE_IPV6)) {
6019 label = htonl(0x2); /* IPV6 Explicit Null. */
6021 label = htonl(0x0); /* IPV4 Explicit Null. */
6023 tc = (ctx->flow.nw_tos & IP_DSCP_MASK) >> 2;
6024 ttl = ctx->flow.nw_ttl ? ctx->flow.nw_ttl : 0x40;
6025 ctx->flow.mpls_lse = set_mpls_lse_values(ttl, tc, 1, label);
6026 ctx->flow.encap_dl_type = ctx->flow.dl_type;
6027 ctx->flow.mpls_depth = 1;
6029 ctx->flow.dl_type = eth_type;
6033 execute_mpls_pop_action(struct action_xlate_ctx *ctx, ovs_be16 eth_type)
6035 ovs_assert(eth_type_mpls(ctx->flow.dl_type));
6036 ovs_assert(!eth_type_mpls(eth_type));
6038 if (ctx->flow.mpls_depth) {
6039 ctx->flow.mpls_depth--;
6040 ctx->flow.mpls_lse = htonl(0);
6041 if (!ctx->flow.mpls_depth) {
6042 ctx->flow.dl_type = eth_type;
6043 ctx->flow.encap_dl_type = htons(0);
6049 compose_dec_ttl(struct action_xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
6051 if (ctx->flow.dl_type != htons(ETH_TYPE_IP) &&
6052 ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) {
6056 if (ctx->flow.nw_ttl > 1) {
6062 for (i = 0; i < ids->n_controllers; i++) {
6063 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
6067 /* Stop processing for current table. */
6073 execute_set_mpls_ttl_action(struct action_xlate_ctx *ctx, uint8_t ttl)
6075 if (!eth_type_mpls(ctx->flow.dl_type)) {
6079 set_mpls_lse_ttl(&ctx->flow.mpls_lse, ttl);
6084 execute_dec_mpls_ttl_action(struct action_xlate_ctx *ctx)
6086 uint8_t ttl = mpls_lse_to_ttl(ctx->flow.mpls_lse);
6088 if (!eth_type_mpls(ctx->flow.dl_type)) {
6094 set_mpls_lse_ttl(&ctx->flow.mpls_lse, ttl);
6097 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
6099 /* Stop processing for current table. */
6105 xlate_output_action(struct action_xlate_ctx *ctx,
6106 uint16_t port, uint16_t max_len, bool may_packet_in)
6108 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
6110 ctx->nf_output_iface = NF_OUT_DROP;
6114 compose_output_action(ctx, ctx->flow.in_port);
6117 xlate_table_action(ctx, ctx->flow.in_port, 0, may_packet_in);
6123 flood_packets(ctx, false);
6126 flood_packets(ctx, true);
6128 case OFPP_CONTROLLER:
6129 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
6135 if (port != ctx->flow.in_port) {
6136 compose_output_action(ctx, port);
6138 xlate_report(ctx, "skipping output to input port");
6143 if (prev_nf_output_iface == NF_OUT_FLOOD) {
6144 ctx->nf_output_iface = NF_OUT_FLOOD;
6145 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
6146 ctx->nf_output_iface = prev_nf_output_iface;
6147 } else if (prev_nf_output_iface != NF_OUT_DROP &&
6148 ctx->nf_output_iface != NF_OUT_FLOOD) {
6149 ctx->nf_output_iface = NF_OUT_MULTI;
6154 xlate_output_reg_action(struct action_xlate_ctx *ctx,
6155 const struct ofpact_output_reg *or)
6157 uint64_t port = mf_get_subfield(&or->src, &ctx->flow);
6158 if (port <= UINT16_MAX) {
6159 xlate_output_action(ctx, port, or->max_len, false);
6164 xlate_enqueue_action(struct action_xlate_ctx *ctx,
6165 const struct ofpact_enqueue *enqueue)
6167 uint16_t ofp_port = enqueue->port;
6168 uint32_t queue_id = enqueue->queue;
6169 uint32_t flow_priority, priority;
6172 /* Translate queue to priority. */
6173 error = dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6174 queue_id, &priority);
6176 /* Fall back to ordinary output action. */
6177 xlate_output_action(ctx, enqueue->port, 0, false);
6181 /* Check output port. */
6182 if (ofp_port == OFPP_IN_PORT) {
6183 ofp_port = ctx->flow.in_port;
6184 } else if (ofp_port == ctx->flow.in_port) {
6188 /* Add datapath actions. */
6189 flow_priority = ctx->flow.skb_priority;
6190 ctx->flow.skb_priority = priority;
6191 compose_output_action(ctx, ofp_port);
6192 ctx->flow.skb_priority = flow_priority;
6194 /* Update NetFlow output port. */
6195 if (ctx->nf_output_iface == NF_OUT_DROP) {
6196 ctx->nf_output_iface = ofp_port;
6197 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
6198 ctx->nf_output_iface = NF_OUT_MULTI;
6203 xlate_set_queue_action(struct action_xlate_ctx *ctx, uint32_t queue_id)
6205 uint32_t skb_priority;
6207 if (!dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6208 queue_id, &skb_priority)) {
6209 ctx->flow.skb_priority = skb_priority;
6211 /* Couldn't translate queue to a priority. Nothing to do. A warning
6212 * has already been logged. */
6216 struct xlate_reg_state {
6222 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
6224 struct ofproto_dpif *ofproto = ofproto_;
6225 struct ofport_dpif *port;
6235 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
6238 port = get_ofp_port(ofproto, ofp_port);
6239 return port ? port->may_enable : false;
6244 xlate_bundle_action(struct action_xlate_ctx *ctx,
6245 const struct ofpact_bundle *bundle)
6249 port = bundle_execute(bundle, &ctx->flow, slave_enabled_cb, ctx->ofproto);
6250 if (bundle->dst.field) {
6251 nxm_reg_load(&bundle->dst, port, &ctx->flow);
6253 xlate_output_action(ctx, port, 0, false);
6258 xlate_learn_action(struct action_xlate_ctx *ctx,
6259 const struct ofpact_learn *learn)
6261 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
6262 struct ofputil_flow_mod fm;
6263 uint64_t ofpacts_stub[1024 / 8];
6264 struct ofpbuf ofpacts;
6267 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
6268 learn_execute(learn, &ctx->flow, &fm, &ofpacts);
6270 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
6271 if (error && !VLOG_DROP_WARN(&rl)) {
6272 VLOG_WARN("learning action failed to modify flow table (%s)",
6273 ofperr_get_name(error));
6276 ofpbuf_uninit(&ofpacts);
6279 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
6280 * means "infinite". */
6282 reduce_timeout(uint16_t max, uint16_t *timeout)
6284 if (max && (!*timeout || *timeout > max)) {
6290 xlate_fin_timeout(struct action_xlate_ctx *ctx,
6291 const struct ofpact_fin_timeout *oft)
6293 if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
6294 struct rule_dpif *rule = ctx->rule;
6296 reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout);
6297 reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout);
6302 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
6304 if (port->up.pp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
6305 ? OFPUTIL_PC_NO_RECV_STP
6306 : OFPUTIL_PC_NO_RECV)) {
6310 /* Only drop packets here if both forwarding and learning are
6311 * disabled. If just learning is enabled, we need to have
6312 * OFPP_NORMAL and the learning action have a look at the packet
6313 * before we can drop it. */
6314 if (!stp_forward_in_state(port->stp_state)
6315 && !stp_learn_in_state(port->stp_state)) {
6323 tunnel_ecn_ok(struct action_xlate_ctx *ctx)
6325 if (is_ip_any(&ctx->base_flow)
6326 && (ctx->base_flow.tunnel.ip_tos & IP_ECN_MASK) == IP_ECN_CE) {
6327 if ((ctx->base_flow.nw_tos & IP_ECN_MASK) == IP_ECN_NOT_ECT) {
6328 VLOG_WARN_RL(&rl, "dropping tunnel packet marked ECN CE"
6329 " but is not ECN capable");
6332 /* Set the ECN CE value in the tunneled packet. */
6333 ctx->flow.nw_tos |= IP_ECN_CE;
6341 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
6342 struct action_xlate_ctx *ctx)
6344 bool was_evictable = true;
6345 const struct ofpact *a;
6348 /* Don't let the rule we're working on get evicted underneath us. */
6349 was_evictable = ctx->rule->up.evictable;
6350 ctx->rule->up.evictable = false;
6352 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
6353 struct ofpact_controller *controller;
6354 const struct ofpact_metadata *metadata;
6362 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
6363 ofpact_get_OUTPUT(a)->max_len, true);
6366 case OFPACT_CONTROLLER:
6367 controller = ofpact_get_CONTROLLER(a);
6368 execute_controller_action(ctx, controller->max_len,
6370 controller->controller_id);
6373 case OFPACT_ENQUEUE:
6374 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
6377 case OFPACT_SET_VLAN_VID:
6378 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
6379 ctx->flow.vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
6383 case OFPACT_SET_VLAN_PCP:
6384 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
6385 ctx->flow.vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp
6390 case OFPACT_STRIP_VLAN:
6391 ctx->flow.vlan_tci = htons(0);
6394 case OFPACT_PUSH_VLAN:
6395 /* XXX 802.1AD(QinQ) */
6396 ctx->flow.vlan_tci = htons(VLAN_CFI);
6399 case OFPACT_SET_ETH_SRC:
6400 memcpy(ctx->flow.dl_src, ofpact_get_SET_ETH_SRC(a)->mac,
6404 case OFPACT_SET_ETH_DST:
6405 memcpy(ctx->flow.dl_dst, ofpact_get_SET_ETH_DST(a)->mac,
6409 case OFPACT_SET_IPV4_SRC:
6410 ctx->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
6413 case OFPACT_SET_IPV4_DST:
6414 ctx->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
6417 case OFPACT_SET_IPV4_DSCP:
6418 /* OpenFlow 1.0 only supports IPv4. */
6419 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
6420 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
6421 ctx->flow.nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
6425 case OFPACT_SET_L4_SRC_PORT:
6426 ctx->flow.tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
6429 case OFPACT_SET_L4_DST_PORT:
6430 ctx->flow.tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
6433 case OFPACT_RESUBMIT:
6434 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
6437 case OFPACT_SET_TUNNEL:
6438 ctx->flow.tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
6441 case OFPACT_SET_QUEUE:
6442 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
6445 case OFPACT_POP_QUEUE:
6446 ctx->flow.skb_priority = ctx->orig_skb_priority;
6449 case OFPACT_REG_MOVE:
6450 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), &ctx->flow);
6453 case OFPACT_REG_LOAD:
6454 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->flow);
6457 case OFPACT_STACK_PUSH:
6458 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), &ctx->flow,
6462 case OFPACT_STACK_POP:
6463 nxm_execute_stack_pop(ofpact_get_STACK_POP(a), &ctx->flow,
6467 case OFPACT_PUSH_MPLS:
6468 execute_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a)->ethertype);
6471 case OFPACT_POP_MPLS:
6472 execute_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype);
6475 case OFPACT_SET_MPLS_TTL:
6476 if (execute_set_mpls_ttl_action(ctx, ofpact_get_SET_MPLS_TTL(a)->ttl)) {
6481 case OFPACT_DEC_MPLS_TTL:
6482 if (execute_dec_mpls_ttl_action(ctx)) {
6487 case OFPACT_DEC_TTL:
6488 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
6494 /* Nothing to do. */
6497 case OFPACT_MULTIPATH:
6498 multipath_execute(ofpact_get_MULTIPATH(a), &ctx->flow);
6502 ctx->ofproto->has_bundle_action = true;
6503 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
6506 case OFPACT_OUTPUT_REG:
6507 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
6511 ctx->has_learn = true;
6512 if (ctx->may_learn) {
6513 xlate_learn_action(ctx, ofpact_get_LEARN(a));
6521 case OFPACT_FIN_TIMEOUT:
6522 ctx->has_fin_timeout = true;
6523 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
6526 case OFPACT_CLEAR_ACTIONS:
6528 * Nothing to do because writa-actions is not supported for now.
6529 * When writa-actions is supported, clear-actions also must
6530 * be supported at the same time.
6534 case OFPACT_WRITE_METADATA:
6535 metadata = ofpact_get_WRITE_METADATA(a);
6536 ctx->flow.metadata &= ~metadata->mask;
6537 ctx->flow.metadata |= metadata->metadata & metadata->mask;
6540 case OFPACT_GOTO_TABLE: {
6541 /* XXX remove recursion */
6542 /* It is assumed that goto-table is last action */
6543 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
6544 ovs_assert(ctx->table_id < ogt->table_id);
6545 xlate_table_action(ctx, ctx->flow.in_port, ogt->table_id, true);
6553 ctx->rule->up.evictable = was_evictable;
6558 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
6559 struct ofproto_dpif *ofproto, const struct flow *flow,
6560 const struct initial_vals *initial_vals,
6561 struct rule_dpif *rule,
6562 uint8_t tcp_flags, const struct ofpbuf *packet)
6564 ovs_be64 initial_tun_id = flow->tunnel.tun_id;
6566 /* Flow initialization rules:
6567 * - 'base_flow' must match the kernel's view of the packet at the
6568 * time that action processing starts. 'flow' represents any
6569 * transformations we wish to make through actions.
6570 * - By default 'base_flow' and 'flow' are the same since the input
6571 * packet matches the output before any actions are applied.
6572 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
6573 * of the received packet as seen by the kernel. If we later output
6574 * to another device without any modifications this will cause us to
6575 * insert a new tag since the original one was stripped off by the
6577 * - Tunnel 'flow' is largely cleared when transitioning between
6578 * the input and output stages since it does not make sense to output
6579 * a packet with the exact headers that it was received with (i.e.
6580 * the destination IP is us). The one exception is the tun_id, which
6581 * is preserved to allow use in later resubmit lookups and loads into
6583 * - Tunnel 'base_flow' is completely cleared since that is what the
6584 * kernel does. If we wish to maintain the original values an action
6585 * needs to be generated. */
6587 ctx->ofproto = ofproto;
6589 memset(&ctx->flow.tunnel, 0, sizeof ctx->flow.tunnel);
6590 ctx->base_flow = ctx->flow;
6591 ctx->base_flow.vlan_tci = initial_vals->vlan_tci;
6592 ctx->base_flow.tunnel.ip_tos = initial_vals->tunnel_ip_tos;
6593 ctx->flow.tunnel.tun_id = initial_tun_id;
6595 ctx->packet = packet;
6596 ctx->may_learn = packet != NULL;
6597 ctx->tcp_flags = tcp_flags;
6598 ctx->resubmit_hook = NULL;
6599 ctx->report_hook = NULL;
6600 ctx->resubmit_stats = NULL;
6603 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
6604 * into datapath actions in 'odp_actions', using 'ctx'. */
6606 xlate_actions(struct action_xlate_ctx *ctx,
6607 const struct ofpact *ofpacts, size_t ofpacts_len,
6608 struct ofpbuf *odp_actions)
6610 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
6611 * that in the future we always keep a copy of the original flow for
6612 * tracing purposes. */
6613 static bool hit_resubmit_limit;
6615 enum slow_path_reason special;
6616 struct ofport_dpif *in_port;
6617 struct flow orig_flow;
6619 COVERAGE_INC(ofproto_dpif_xlate);
6621 ofpbuf_clear(odp_actions);
6622 ofpbuf_reserve(odp_actions, NL_A_U32_SIZE);
6624 ctx->odp_actions = odp_actions;
6627 ctx->has_learn = false;
6628 ctx->has_normal = false;
6629 ctx->has_fin_timeout = false;
6630 ctx->nf_output_iface = NF_OUT_DROP;
6633 ctx->max_resubmit_trigger = false;
6634 ctx->orig_skb_priority = ctx->flow.skb_priority;
6638 ofpbuf_use_stub(&ctx->stack, ctx->init_stack, sizeof ctx->init_stack);
6640 if (ctx->ofproto->has_mirrors || hit_resubmit_limit) {
6641 /* Do this conditionally because the copy is expensive enough that it
6642 * shows up in profiles. */
6643 orig_flow = ctx->flow;
6646 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
6647 switch (ctx->ofproto->up.frag_handling) {
6648 case OFPC_FRAG_NORMAL:
6649 /* We must pretend that transport ports are unavailable. */
6650 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
6651 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
6654 case OFPC_FRAG_DROP:
6657 case OFPC_FRAG_REASM:
6660 case OFPC_FRAG_NX_MATCH:
6661 /* Nothing to do. */
6664 case OFPC_INVALID_TTL_TO_CONTROLLER:
6669 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
6670 special = process_special(ctx->ofproto, &ctx->flow, in_port, ctx->packet);
6672 ctx->slow |= special;
6674 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
6675 struct initial_vals initial_vals;
6676 uint32_t local_odp_port;
6678 initial_vals.vlan_tci = ctx->base_flow.vlan_tci;
6679 initial_vals.tunnel_ip_tos = ctx->base_flow.tunnel.ip_tos;
6681 add_sflow_action(ctx);
6683 if (tunnel_ecn_ok(ctx) && (!in_port || may_receive(in_port, ctx))) {
6684 do_xlate_actions(ofpacts, ofpacts_len, ctx);
6686 /* We've let OFPP_NORMAL and the learning action look at the
6687 * packet, so drop it now if forwarding is disabled. */
6688 if (in_port && !stp_forward_in_state(in_port->stp_state)) {
6689 ofpbuf_clear(ctx->odp_actions);
6690 add_sflow_action(ctx);
6694 if (ctx->max_resubmit_trigger && !ctx->resubmit_hook) {
6695 if (!hit_resubmit_limit) {
6696 /* We didn't record the original flow. Make sure we do from
6698 hit_resubmit_limit = true;
6699 } else if (!VLOG_DROP_ERR(&trace_rl)) {
6700 struct ds ds = DS_EMPTY_INITIALIZER;
6702 ofproto_trace(ctx->ofproto, &orig_flow, ctx->packet,
6703 &initial_vals, &ds);
6704 VLOG_ERR("Trace triggered by excessive resubmit "
6705 "recursion:\n%s", ds_cstr(&ds));
6710 local_odp_port = ofp_port_to_odp_port(ctx->ofproto, OFPP_LOCAL);
6711 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
6713 ctx->odp_actions->data,
6714 ctx->odp_actions->size)) {
6715 ctx->slow |= SLOW_IN_BAND;
6717 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
6719 compose_output_action(ctx, OFPP_LOCAL);
6722 if (ctx->ofproto->has_mirrors) {
6723 add_mirror_actions(ctx, &orig_flow);
6725 fix_sflow_action(ctx);
6728 ofpbuf_uninit(&ctx->stack);
6731 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
6732 * into datapath actions, using 'ctx', and discards the datapath actions. */
6734 xlate_actions_for_side_effects(struct action_xlate_ctx *ctx,
6735 const struct ofpact *ofpacts,
6738 uint64_t odp_actions_stub[1024 / 8];
6739 struct ofpbuf odp_actions;
6741 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
6742 xlate_actions(ctx, ofpacts, ofpacts_len, &odp_actions);
6743 ofpbuf_uninit(&odp_actions);
6747 xlate_report(struct action_xlate_ctx *ctx, const char *s)
6749 if (ctx->report_hook) {
6750 ctx->report_hook(ctx, s);
6754 /* OFPP_NORMAL implementation. */
6756 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
6758 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
6759 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
6760 * the bundle on which the packet was received, returns the VLAN to which the
6763 * Both 'vid' and the return value are in the range 0...4095. */
6765 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
6767 switch (in_bundle->vlan_mode) {
6768 case PORT_VLAN_ACCESS:
6769 return in_bundle->vlan;
6772 case PORT_VLAN_TRUNK:
6775 case PORT_VLAN_NATIVE_UNTAGGED:
6776 case PORT_VLAN_NATIVE_TAGGED:
6777 return vid ? vid : in_bundle->vlan;
6784 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
6785 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
6788 * 'vid' should be the VID obtained from the 802.1Q header that was received as
6789 * part of a packet (specify 0 if there was no 802.1Q header), in the range
6792 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
6794 /* Allow any VID on the OFPP_NONE port. */
6795 if (in_bundle == &ofpp_none_bundle) {
6799 switch (in_bundle->vlan_mode) {
6800 case PORT_VLAN_ACCESS:
6803 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6804 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
6805 "packet received on port %s configured as VLAN "
6806 "%"PRIu16" access port",
6807 in_bundle->ofproto->up.name, vid,
6808 in_bundle->name, in_bundle->vlan);
6814 case PORT_VLAN_NATIVE_UNTAGGED:
6815 case PORT_VLAN_NATIVE_TAGGED:
6817 /* Port must always carry its native VLAN. */
6821 case PORT_VLAN_TRUNK:
6822 if (!ofbundle_includes_vlan(in_bundle, vid)) {
6824 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6825 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
6826 "received on port %s not configured for trunking "
6828 in_bundle->ofproto->up.name, vid,
6829 in_bundle->name, vid);
6841 /* Given 'vlan', the VLAN that a packet belongs to, and
6842 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
6843 * that should be included in the 802.1Q header. (If the return value is 0,
6844 * then the 802.1Q header should only be included in the packet if there is a
6847 * Both 'vlan' and the return value are in the range 0...4095. */
6849 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
6851 switch (out_bundle->vlan_mode) {
6852 case PORT_VLAN_ACCESS:
6855 case PORT_VLAN_TRUNK:
6856 case PORT_VLAN_NATIVE_TAGGED:
6859 case PORT_VLAN_NATIVE_UNTAGGED:
6860 return vlan == out_bundle->vlan ? 0 : vlan;
6868 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
6871 struct ofport_dpif *port;
6873 ovs_be16 tci, old_tci;
6875 vid = output_vlan_to_vid(out_bundle, vlan);
6876 if (!out_bundle->bond) {
6877 port = ofbundle_get_a_port(out_bundle);
6879 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
6882 /* No slaves enabled, so drop packet. */
6887 old_tci = ctx->flow.vlan_tci;
6889 if (tci || out_bundle->use_priority_tags) {
6890 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
6892 tci |= htons(VLAN_CFI);
6895 ctx->flow.vlan_tci = tci;
6897 compose_output_action(ctx, port->up.ofp_port);
6898 ctx->flow.vlan_tci = old_tci;
6902 mirror_mask_ffs(mirror_mask_t mask)
6904 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
6909 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
6911 return (bundle->vlan_mode != PORT_VLAN_ACCESS
6912 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
6916 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
6918 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
6921 /* Returns an arbitrary interface within 'bundle'. */
6922 static struct ofport_dpif *
6923 ofbundle_get_a_port(const struct ofbundle *bundle)
6925 return CONTAINER_OF(list_front(&bundle->ports),
6926 struct ofport_dpif, bundle_node);
6930 vlan_is_mirrored(const struct ofmirror *m, int vlan)
6932 return !m->vlans || bitmap_is_set(m->vlans, vlan);
6936 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
6938 struct ofproto_dpif *ofproto = ctx->ofproto;
6939 mirror_mask_t mirrors;
6940 struct ofbundle *in_bundle;
6943 const struct nlattr *a;
6946 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
6947 ctx->packet != NULL, NULL);
6951 mirrors = in_bundle->src_mirrors;
6953 /* Drop frames on bundles reserved for mirroring. */
6954 if (in_bundle->mirror_out) {
6955 if (ctx->packet != NULL) {
6956 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6957 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
6958 "%s, which is reserved exclusively for mirroring",
6959 ctx->ofproto->up.name, in_bundle->name);
6965 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
6966 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
6969 vlan = input_vid_to_vlan(in_bundle, vid);
6971 /* Look at the output ports to check for destination selections. */
6973 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
6974 ctx->odp_actions->size) {
6975 enum ovs_action_attr type = nl_attr_type(a);
6976 struct ofport_dpif *ofport;
6978 if (type != OVS_ACTION_ATTR_OUTPUT) {
6982 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
6983 if (ofport && ofport->bundle) {
6984 mirrors |= ofport->bundle->dst_mirrors;
6992 /* Restore the original packet before adding the mirror actions. */
6993 ctx->flow = *orig_flow;
6998 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
7000 if (!vlan_is_mirrored(m, vlan)) {
7001 mirrors = zero_rightmost_1bit(mirrors);
7005 mirrors &= ~m->dup_mirrors;
7006 ctx->mirrors |= m->dup_mirrors;
7008 output_normal(ctx, m->out, vlan);
7009 } else if (vlan != m->out_vlan
7010 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
7011 struct ofbundle *bundle;
7013 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
7014 if (ofbundle_includes_vlan(bundle, m->out_vlan)
7015 && !bundle->mirror_out) {
7016 output_normal(ctx, bundle, m->out_vlan);
7024 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
7025 uint64_t packets, uint64_t bytes)
7031 for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) {
7034 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
7037 /* In normal circumstances 'm' will not be NULL. However,
7038 * if mirrors are reconfigured, we can temporarily get out
7039 * of sync in facet_revalidate(). We could "correct" the
7040 * mirror list before reaching here, but doing that would
7041 * not properly account the traffic stats we've currently
7042 * accumulated for previous mirror configuration. */
7046 m->packet_count += packets;
7047 m->byte_count += bytes;
7051 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
7052 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
7053 * indicate this; newer upstream kernels use gratuitous ARP requests. */
7055 is_gratuitous_arp(const struct flow *flow)
7057 return (flow->dl_type == htons(ETH_TYPE_ARP)
7058 && eth_addr_is_broadcast(flow->dl_dst)
7059 && (flow->nw_proto == ARP_OP_REPLY
7060 || (flow->nw_proto == ARP_OP_REQUEST
7061 && flow->nw_src == flow->nw_dst)));
7065 update_learning_table(struct ofproto_dpif *ofproto,
7066 const struct flow *flow, int vlan,
7067 struct ofbundle *in_bundle)
7069 struct mac_entry *mac;
7071 /* Don't learn the OFPP_NONE port. */
7072 if (in_bundle == &ofpp_none_bundle) {
7076 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
7080 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
7081 if (is_gratuitous_arp(flow)) {
7082 /* We don't want to learn from gratuitous ARP packets that are
7083 * reflected back over bond slaves so we lock the learning table. */
7084 if (!in_bundle->bond) {
7085 mac_entry_set_grat_arp_lock(mac);
7086 } else if (mac_entry_is_grat_arp_locked(mac)) {
7091 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
7092 /* The log messages here could actually be useful in debugging,
7093 * so keep the rate limit relatively high. */
7094 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
7095 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
7096 "on port %s in VLAN %d",
7097 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
7098 in_bundle->name, vlan);
7100 mac->port.p = in_bundle;
7101 tag_set_add(&ofproto->backer->revalidate_set,
7102 mac_learning_changed(ofproto->ml, mac));
7106 static struct ofbundle *
7107 lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port,
7108 bool warn, struct ofport_dpif **in_ofportp)
7110 struct ofport_dpif *ofport;
7112 /* Find the port and bundle for the received packet. */
7113 ofport = get_ofp_port(ofproto, in_port);
7115 *in_ofportp = ofport;
7117 if (ofport && ofport->bundle) {
7118 return ofport->bundle;
7121 /* Special-case OFPP_NONE, which a controller may use as the ingress
7122 * port for traffic that it is sourcing. */
7123 if (in_port == OFPP_NONE) {
7124 return &ofpp_none_bundle;
7127 /* Odd. A few possible reasons here:
7129 * - We deleted a port but there are still a few packets queued up
7132 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
7133 * we don't know about.
7135 * - The ofproto client didn't configure the port as part of a bundle.
7136 * This is particularly likely to happen if a packet was received on the
7137 * port after it was created, but before the client had a chance to
7138 * configure its bundle.
7141 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7143 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
7144 "port %"PRIu16, ofproto->up.name, in_port);
7149 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
7150 * dropped. Returns true if they may be forwarded, false if they should be
7153 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
7154 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
7156 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
7157 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
7158 * checked by input_vid_is_valid().
7160 * May also add tags to '*tags', although the current implementation only does
7161 * so in one special case.
7164 is_admissible(struct action_xlate_ctx *ctx, struct ofport_dpif *in_port,
7167 struct ofproto_dpif *ofproto = ctx->ofproto;
7168 struct flow *flow = &ctx->flow;
7169 struct ofbundle *in_bundle = in_port->bundle;
7171 /* Drop frames for reserved multicast addresses
7172 * only if forward_bpdu option is absent. */
7173 if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
7174 xlate_report(ctx, "packet has reserved destination MAC, dropping");
7178 if (in_bundle->bond) {
7179 struct mac_entry *mac;
7181 switch (bond_check_admissibility(in_bundle->bond, in_port,
7182 flow->dl_dst, &ctx->tags)) {
7187 xlate_report(ctx, "bonding refused admissibility, dropping");
7190 case BV_DROP_IF_MOVED:
7191 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
7192 if (mac && mac->port.p != in_bundle &&
7193 (!is_gratuitous_arp(flow)
7194 || mac_entry_is_grat_arp_locked(mac))) {
7195 xlate_report(ctx, "SLB bond thinks this packet looped back, "
7207 xlate_normal(struct action_xlate_ctx *ctx)
7209 struct ofport_dpif *in_port;
7210 struct ofbundle *in_bundle;
7211 struct mac_entry *mac;
7215 ctx->has_normal = true;
7217 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
7218 ctx->packet != NULL, &in_port);
7220 xlate_report(ctx, "no input bundle, dropping");
7224 /* Drop malformed frames. */
7225 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
7226 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
7227 if (ctx->packet != NULL) {
7228 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7229 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
7230 "VLAN tag received on port %s",
7231 ctx->ofproto->up.name, in_bundle->name);
7233 xlate_report(ctx, "partial VLAN tag, dropping");
7237 /* Drop frames on bundles reserved for mirroring. */
7238 if (in_bundle->mirror_out) {
7239 if (ctx->packet != NULL) {
7240 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7241 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7242 "%s, which is reserved exclusively for mirroring",
7243 ctx->ofproto->up.name, in_bundle->name);
7245 xlate_report(ctx, "input port is mirror output port, dropping");
7250 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
7251 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
7252 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
7255 vlan = input_vid_to_vlan(in_bundle, vid);
7257 /* Check other admissibility requirements. */
7258 if (in_port && !is_admissible(ctx, in_port, vlan)) {
7262 /* Learn source MAC. */
7263 if (ctx->may_learn) {
7264 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
7267 /* Determine output bundle. */
7268 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
7271 if (mac->port.p != in_bundle) {
7272 xlate_report(ctx, "forwarding to learned port");
7273 output_normal(ctx, mac->port.p, vlan);
7275 xlate_report(ctx, "learned port is input port, dropping");
7278 struct ofbundle *bundle;
7280 xlate_report(ctx, "no learned MAC for destination, flooding");
7281 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
7282 if (bundle != in_bundle
7283 && ofbundle_includes_vlan(bundle, vlan)
7284 && bundle->floodable
7285 && !bundle->mirror_out) {
7286 output_normal(ctx, bundle, vlan);
7289 ctx->nf_output_iface = NF_OUT_FLOOD;
7293 /* Optimized flow revalidation.
7295 * It's a difficult problem, in general, to tell which facets need to have
7296 * their actions recalculated whenever the OpenFlow flow table changes. We
7297 * don't try to solve that general problem: for most kinds of OpenFlow flow
7298 * table changes, we recalculate the actions for every facet. This is
7299 * relatively expensive, but it's good enough if the OpenFlow flow table
7300 * doesn't change very often.
7302 * However, we can expect one particular kind of OpenFlow flow table change to
7303 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
7304 * of CPU on revalidating every facet whenever MAC learning modifies the flow
7305 * table, we add a special case that applies to flow tables in which every rule
7306 * has the same form (that is, the same wildcards), except that the table is
7307 * also allowed to have a single "catch-all" flow that matches all packets. We
7308 * optimize this case by tagging all of the facets that resubmit into the table
7309 * and invalidating the same tag whenever a flow changes in that table. The
7310 * end result is that we revalidate just the facets that need it (and sometimes
7311 * a few more, but not all of the facets or even all of the facets that
7312 * resubmit to the table modified by MAC learning). */
7314 /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted
7315 * into an OpenFlow table with the given 'basis'. */
7317 rule_calculate_tag(const struct flow *flow, const struct minimask *mask,
7320 if (minimask_is_catchall(mask)) {
7323 uint32_t hash = flow_hash_in_minimask(flow, mask, secret);
7324 return tag_create_deterministic(hash);
7328 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
7329 * taggability of that table.
7331 * This function must be called after *each* change to a flow table. If you
7332 * skip calling it on some changes then the pointer comparisons at the end can
7333 * be invalid if you get unlucky. For example, if a flow removal causes a
7334 * cls_table to be destroyed and then a flow insertion causes a cls_table with
7335 * different wildcards to be created with the same address, then this function
7336 * will incorrectly skip revalidation. */
7338 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
7340 struct table_dpif *table = &ofproto->tables[table_id];
7341 const struct oftable *oftable = &ofproto->up.tables[table_id];
7342 struct cls_table *catchall, *other;
7343 struct cls_table *t;
7345 catchall = other = NULL;
7347 switch (hmap_count(&oftable->cls.tables)) {
7349 /* We could tag this OpenFlow table but it would make the logic a
7350 * little harder and it's a corner case that doesn't seem worth it
7356 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
7357 if (cls_table_is_catchall(t)) {
7359 } else if (!other) {
7362 /* Indicate that we can't tag this by setting both tables to
7363 * NULL. (We know that 'catchall' is already NULL.) */
7370 /* Can't tag this table. */
7374 if (table->catchall_table != catchall || table->other_table != other) {
7375 table->catchall_table = catchall;
7376 table->other_table = other;
7377 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7381 /* Given 'rule' that has changed in some way (either it is a rule being
7382 * inserted, a rule being deleted, or a rule whose actions are being
7383 * modified), marks facets for revalidation to ensure that packets will be
7384 * forwarded correctly according to the new state of the flow table.
7386 * This function must be called after *each* change to a flow table. See
7387 * the comment on table_update_taggable() for more information. */
7389 rule_invalidate(const struct rule_dpif *rule)
7391 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
7393 table_update_taggable(ofproto, rule->up.table_id);
7395 if (!ofproto->backer->need_revalidate) {
7396 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
7398 if (table->other_table && rule->tag) {
7399 tag_set_add(&ofproto->backer->revalidate_set, rule->tag);
7401 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7407 set_frag_handling(struct ofproto *ofproto_,
7408 enum ofp_config_flags frag_handling)
7410 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7411 if (frag_handling != OFPC_FRAG_REASM) {
7412 ofproto->backer->need_revalidate = REV_RECONFIGURE;
7420 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
7421 const struct flow *flow,
7422 const struct ofpact *ofpacts, size_t ofpacts_len)
7424 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7425 struct initial_vals initial_vals;
7426 struct odputil_keybuf keybuf;
7427 struct dpif_flow_stats stats;
7431 struct action_xlate_ctx ctx;
7432 uint64_t odp_actions_stub[1024 / 8];
7433 struct ofpbuf odp_actions;
7435 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
7436 odp_flow_key_from_flow(&key, flow,
7437 ofp_port_to_odp_port(ofproto, flow->in_port));
7439 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
7441 initial_vals.vlan_tci = flow->vlan_tci;
7442 initial_vals.tunnel_ip_tos = 0;
7443 action_xlate_ctx_init(&ctx, ofproto, flow, &initial_vals, NULL,
7444 packet_get_tcp_flags(packet, flow), packet);
7445 ctx.resubmit_stats = &stats;
7447 ofpbuf_use_stub(&odp_actions,
7448 odp_actions_stub, sizeof odp_actions_stub);
7449 xlate_actions(&ctx, ofpacts, ofpacts_len, &odp_actions);
7450 dpif_execute(ofproto->backer->dpif, key.data, key.size,
7451 odp_actions.data, odp_actions.size, packet);
7452 ofpbuf_uninit(&odp_actions);
7460 set_netflow(struct ofproto *ofproto_,
7461 const struct netflow_options *netflow_options)
7463 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7465 if (netflow_options) {
7466 if (!ofproto->netflow) {
7467 ofproto->netflow = netflow_create();
7469 return netflow_set_options(ofproto->netflow, netflow_options);
7471 netflow_destroy(ofproto->netflow);
7472 ofproto->netflow = NULL;
7478 get_netflow_ids(const struct ofproto *ofproto_,
7479 uint8_t *engine_type, uint8_t *engine_id)
7481 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7483 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
7487 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
7489 if (!facet_is_controller_flow(facet) &&
7490 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
7491 struct subfacet *subfacet;
7492 struct ofexpired expired;
7494 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
7495 if (subfacet->path == SF_FAST_PATH) {
7496 struct dpif_flow_stats stats;
7498 subfacet_reinstall(subfacet, &stats);
7499 subfacet_update_stats(subfacet, &stats);
7503 expired.flow = facet->flow;
7504 expired.packet_count = facet->packet_count;
7505 expired.byte_count = facet->byte_count;
7506 expired.used = facet->used;
7507 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
7512 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
7514 struct facet *facet;
7516 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
7517 send_active_timeout(ofproto, facet);
7521 static struct ofproto_dpif *
7522 ofproto_dpif_lookup(const char *name)
7524 struct ofproto_dpif *ofproto;
7526 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
7527 hash_string(name, 0), &all_ofproto_dpifs) {
7528 if (!strcmp(ofproto->up.name, name)) {
7536 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
7537 const char *argv[], void *aux OVS_UNUSED)
7539 struct ofproto_dpif *ofproto;
7542 ofproto = ofproto_dpif_lookup(argv[1]);
7544 unixctl_command_reply_error(conn, "no such bridge");
7547 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
7549 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7550 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
7554 unixctl_command_reply(conn, "table successfully flushed");
7558 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
7559 const char *argv[], void *aux OVS_UNUSED)
7561 struct ds ds = DS_EMPTY_INITIALIZER;
7562 const struct ofproto_dpif *ofproto;
7563 const struct mac_entry *e;
7565 ofproto = ofproto_dpif_lookup(argv[1]);
7567 unixctl_command_reply_error(conn, "no such bridge");
7571 ds_put_cstr(&ds, " port VLAN MAC Age\n");
7572 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
7573 struct ofbundle *bundle = e->port.p;
7574 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
7575 ofbundle_get_a_port(bundle)->odp_port,
7576 e->vlan, ETH_ADDR_ARGS(e->mac),
7577 mac_entry_age(ofproto->ml, e));
7579 unixctl_command_reply(conn, ds_cstr(&ds));
7584 struct action_xlate_ctx ctx;
7590 trace_format_rule(struct ds *result, uint8_t table_id, int level,
7591 const struct rule_dpif *rule)
7593 ds_put_char_multiple(result, '\t', level);
7595 ds_put_cstr(result, "No match\n");
7599 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
7600 table_id, ntohll(rule->up.flow_cookie));
7601 cls_rule_format(&rule->up.cr, result);
7602 ds_put_char(result, '\n');
7604 ds_put_char_multiple(result, '\t', level);
7605 ds_put_cstr(result, "OpenFlow ");
7606 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
7607 ds_put_char(result, '\n');
7611 trace_format_flow(struct ds *result, int level, const char *title,
7612 struct trace_ctx *trace)
7614 ds_put_char_multiple(result, '\t', level);
7615 ds_put_format(result, "%s: ", title);
7616 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
7617 ds_put_cstr(result, "unchanged");
7619 flow_format(result, &trace->ctx.flow);
7620 trace->flow = trace->ctx.flow;
7622 ds_put_char(result, '\n');
7626 trace_format_regs(struct ds *result, int level, const char *title,
7627 struct trace_ctx *trace)
7631 ds_put_char_multiple(result, '\t', level);
7632 ds_put_format(result, "%s:", title);
7633 for (i = 0; i < FLOW_N_REGS; i++) {
7634 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
7636 ds_put_char(result, '\n');
7640 trace_format_odp(struct ds *result, int level, const char *title,
7641 struct trace_ctx *trace)
7643 struct ofpbuf *odp_actions = trace->ctx.odp_actions;
7645 ds_put_char_multiple(result, '\t', level);
7646 ds_put_format(result, "%s: ", title);
7647 format_odp_actions(result, odp_actions->data, odp_actions->size);
7648 ds_put_char(result, '\n');
7652 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
7654 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
7655 struct ds *result = trace->result;
7657 ds_put_char(result, '\n');
7658 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
7659 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
7660 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
7661 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
7665 trace_report(struct action_xlate_ctx *ctx, const char *s)
7667 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
7668 struct ds *result = trace->result;
7670 ds_put_char_multiple(result, '\t', ctx->recurse);
7671 ds_put_cstr(result, s);
7672 ds_put_char(result, '\n');
7676 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
7677 void *aux OVS_UNUSED)
7679 const char *dpname = argv[1];
7680 struct ofproto_dpif *ofproto;
7681 struct ofpbuf odp_key;
7682 struct ofpbuf *packet;
7683 struct initial_vals initial_vals;
7689 ofpbuf_init(&odp_key, 0);
7692 ofproto = ofproto_dpif_lookup(dpname);
7694 unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list "
7698 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
7699 /* ofproto/trace dpname flow [-generate] */
7700 const char *flow_s = argv[2];
7701 const char *generate_s = argv[3];
7703 /* Allow 'flow_s' to be either a datapath flow or an OpenFlow-like
7704 * flow. We guess which type it is based on whether 'flow_s' contains
7705 * an '(', since a datapath flow always contains '(') but an
7706 * OpenFlow-like flow should not (in fact it's allowed but I believe
7707 * that's not documented anywhere).
7709 * An alternative would be to try to parse 'flow_s' both ways, but then
7710 * it would be tricky giving a sensible error message. After all, do
7711 * you just say "syntax error" or do you present both error messages?
7712 * Both choices seem lousy. */
7713 if (strchr(flow_s, '(')) {
7716 /* Convert string to datapath key. */
7717 ofpbuf_init(&odp_key, 0);
7718 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
7720 unixctl_command_reply_error(conn, "Bad flow syntax");
7724 /* XXX: Since we allow the user to specify an ofproto, it's
7725 * possible they will specify a different ofproto than the one the
7726 * port actually belongs too. Ideally we should simply remove the
7727 * ability to specify the ofproto. */
7728 if (ofproto_receive(ofproto->backer, NULL, odp_key.data,
7729 odp_key.size, &flow, NULL, NULL, NULL,
7731 unixctl_command_reply_error(conn, "Invalid flow");
7737 error_s = parse_ofp_exact_flow(&flow, argv[2]);
7739 unixctl_command_reply_error(conn, error_s);
7744 initial_vals.vlan_tci = flow.vlan_tci;
7745 initial_vals.tunnel_ip_tos = flow.tunnel.ip_tos;
7748 /* Generate a packet, if requested. */
7750 packet = ofpbuf_new(0);
7751 flow_compose(packet, &flow);
7753 } else if (argc == 7) {
7754 /* ofproto/trace dpname priority tun_id in_port mark packet */
7755 const char *priority_s = argv[2];
7756 const char *tun_id_s = argv[3];
7757 const char *in_port_s = argv[4];
7758 const char *mark_s = argv[5];
7759 const char *packet_s = argv[6];
7760 uint32_t in_port = atoi(in_port_s);
7761 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
7762 uint32_t priority = atoi(priority_s);
7763 uint32_t mark = atoi(mark_s);
7766 msg = eth_from_hex(packet_s, &packet);
7768 unixctl_command_reply_error(conn, msg);
7772 ds_put_cstr(&result, "Packet: ");
7773 s = ofp_packet_to_string(packet->data, packet->size);
7774 ds_put_cstr(&result, s);
7777 flow_extract(packet, priority, mark, NULL, in_port, &flow);
7778 flow.tunnel.tun_id = tun_id;
7779 initial_vals.vlan_tci = flow.vlan_tci;
7780 initial_vals.tunnel_ip_tos = flow.tunnel.ip_tos;
7782 unixctl_command_reply_error(conn, "Bad command syntax");
7786 ofproto_trace(ofproto, &flow, packet, &initial_vals, &result);
7787 unixctl_command_reply(conn, ds_cstr(&result));
7790 ds_destroy(&result);
7791 ofpbuf_delete(packet);
7792 ofpbuf_uninit(&odp_key);
7796 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
7797 const struct ofpbuf *packet,
7798 const struct initial_vals *initial_vals, struct ds *ds)
7800 struct rule_dpif *rule;
7802 ds_put_cstr(ds, "Flow: ");
7803 flow_format(ds, flow);
7804 ds_put_char(ds, '\n');
7806 rule = rule_dpif_lookup(ofproto, flow);
7808 trace_format_rule(ds, 0, 0, rule);
7809 if (rule == ofproto->miss_rule) {
7810 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
7811 } else if (rule == ofproto->no_packet_in_rule) {
7812 ds_put_cstr(ds, "\nNo match, packets dropped because "
7813 "OFPPC_NO_PACKET_IN is set on in_port.\n");
7817 uint64_t odp_actions_stub[1024 / 8];
7818 struct ofpbuf odp_actions;
7820 struct trace_ctx trace;
7823 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
7826 ofpbuf_use_stub(&odp_actions,
7827 odp_actions_stub, sizeof odp_actions_stub);
7828 action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_vals,
7829 rule, tcp_flags, packet);
7830 trace.ctx.resubmit_hook = trace_resubmit;
7831 trace.ctx.report_hook = trace_report;
7832 xlate_actions(&trace.ctx, rule->up.ofpacts, rule->up.ofpacts_len,
7835 ds_put_char(ds, '\n');
7836 trace_format_flow(ds, 0, "Final flow", &trace);
7837 ds_put_cstr(ds, "Datapath actions: ");
7838 format_odp_actions(ds, odp_actions.data, odp_actions.size);
7839 ofpbuf_uninit(&odp_actions);
7841 if (trace.ctx.slow) {
7842 enum slow_path_reason slow;
7844 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
7845 "slow path because it:");
7846 for (slow = trace.ctx.slow; slow; ) {
7847 enum slow_path_reason bit = rightmost_1bit(slow);
7851 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
7854 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
7857 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
7860 ds_put_cstr(ds, "\n\t- Needs in-band special case "
7863 ds_put_cstr(ds, "\n\t (The datapath actions are "
7864 "incomplete--for complete actions, "
7865 "please supply a packet.)");
7868 case SLOW_CONTROLLER:
7869 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
7870 "to the OpenFlow controller.");
7873 ds_put_cstr(ds, "\n\t- Needs more specific matching "
7874 "than the datapath supports.");
7881 if (slow & ~SLOW_MATCH) {
7882 ds_put_cstr(ds, "\nThe datapath actions above do not reflect "
7883 "the special slow-path processing.");
7890 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
7891 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
7894 unixctl_command_reply(conn, NULL);
7898 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
7899 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
7902 unixctl_command_reply(conn, NULL);
7905 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
7906 * 'reply' describing the results. */
7908 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
7910 struct facet *facet;
7914 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
7915 if (!facet_check_consistency(facet)) {
7920 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
7924 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
7925 ofproto->up.name, errors);
7927 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
7932 ofproto_dpif_self_check(struct unixctl_conn *conn,
7933 int argc, const char *argv[], void *aux OVS_UNUSED)
7935 struct ds reply = DS_EMPTY_INITIALIZER;
7936 struct ofproto_dpif *ofproto;
7939 ofproto = ofproto_dpif_lookup(argv[1]);
7941 unixctl_command_reply_error(conn, "Unknown ofproto (use "
7942 "ofproto/list for help)");
7945 ofproto_dpif_self_check__(ofproto, &reply);
7947 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7948 ofproto_dpif_self_check__(ofproto, &reply);
7952 unixctl_command_reply(conn, ds_cstr(&reply));
7956 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
7957 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
7958 * to destroy 'ofproto_shash' and free the returned value. */
7959 static const struct shash_node **
7960 get_ofprotos(struct shash *ofproto_shash)
7962 const struct ofproto_dpif *ofproto;
7964 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7965 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
7966 shash_add_nocopy(ofproto_shash, name, ofproto);
7969 return shash_sort(ofproto_shash);
7973 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
7974 const char *argv[] OVS_UNUSED,
7975 void *aux OVS_UNUSED)
7977 struct ds ds = DS_EMPTY_INITIALIZER;
7978 struct shash ofproto_shash;
7979 const struct shash_node **sorted_ofprotos;
7982 shash_init(&ofproto_shash);
7983 sorted_ofprotos = get_ofprotos(&ofproto_shash);
7984 for (i = 0; i < shash_count(&ofproto_shash); i++) {
7985 const struct shash_node *node = sorted_ofprotos[i];
7986 ds_put_format(&ds, "%s\n", node->name);
7989 shash_destroy(&ofproto_shash);
7990 free(sorted_ofprotos);
7992 unixctl_command_reply(conn, ds_cstr(&ds));
7997 show_dp_format(const struct ofproto_dpif *ofproto, struct ds *ds)
7999 struct dpif_dp_stats s;
8000 const struct shash_node **ports;
8003 dpif_get_dp_stats(ofproto->backer->dpif, &s);
8005 ds_put_format(ds, "%s (%s):\n", ofproto->up.name,
8006 dpif_name(ofproto->backer->dpif));
8007 /* xxx It would be better to show bridge-specific stats instead
8008 * xxx of dp ones. */
8010 "\tlookups: hit:%"PRIu64" missed:%"PRIu64" lost:%"PRIu64"\n",
8011 s.n_hit, s.n_missed, s.n_lost);
8012 ds_put_format(ds, "\tflows: %zu\n",
8013 hmap_count(&ofproto->subfacets));
8015 ports = shash_sort(&ofproto->up.port_by_name);
8016 for (i = 0; i < shash_count(&ofproto->up.port_by_name); i++) {
8017 const struct shash_node *node = ports[i];
8018 struct ofport *ofport = node->data;
8019 const char *name = netdev_get_name(ofport->netdev);
8020 const char *type = netdev_get_type(ofport->netdev);
8023 ds_put_format(ds, "\t%s %u/", name, ofport->ofp_port);
8025 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
8026 if (odp_port != OVSP_NONE) {
8027 ds_put_format(ds, "%"PRIu32":", odp_port);
8029 ds_put_cstr(ds, "none:");
8032 if (strcmp(type, "system")) {
8033 struct netdev *netdev;
8036 ds_put_format(ds, " (%s", type);
8038 error = netdev_open(name, type, &netdev);
8043 error = netdev_get_config(netdev, &config);
8045 const struct smap_node **nodes;
8048 nodes = smap_sort(&config);
8049 for (i = 0; i < smap_count(&config); i++) {
8050 const struct smap_node *node = nodes[i];
8051 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
8052 node->key, node->value);
8056 smap_destroy(&config);
8058 netdev_close(netdev);
8060 ds_put_char(ds, ')');
8062 ds_put_char(ds, '\n');
8068 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc,
8069 const char *argv[], void *aux OVS_UNUSED)
8071 struct ds ds = DS_EMPTY_INITIALIZER;
8072 const struct ofproto_dpif *ofproto;
8076 for (i = 1; i < argc; i++) {
8077 ofproto = ofproto_dpif_lookup(argv[i]);
8079 ds_put_format(&ds, "Unknown bridge %s (use dpif/dump-dps "
8080 "for help)", argv[i]);
8081 unixctl_command_reply_error(conn, ds_cstr(&ds));
8084 show_dp_format(ofproto, &ds);
8087 struct shash ofproto_shash;
8088 const struct shash_node **sorted_ofprotos;
8091 shash_init(&ofproto_shash);
8092 sorted_ofprotos = get_ofprotos(&ofproto_shash);
8093 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8094 const struct shash_node *node = sorted_ofprotos[i];
8095 show_dp_format(node->data, &ds);
8098 shash_destroy(&ofproto_shash);
8099 free(sorted_ofprotos);
8102 unixctl_command_reply(conn, ds_cstr(&ds));
8107 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
8108 int argc OVS_UNUSED, const char *argv[],
8109 void *aux OVS_UNUSED)
8111 struct ds ds = DS_EMPTY_INITIALIZER;
8112 const struct ofproto_dpif *ofproto;
8113 struct subfacet *subfacet;
8115 ofproto = ofproto_dpif_lookup(argv[1]);
8117 unixctl_command_reply_error(conn, "no such bridge");
8121 update_stats(ofproto->backer);
8123 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
8124 odp_flow_key_format(subfacet->key, subfacet->key_len, &ds);
8126 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
8127 subfacet->dp_packet_count, subfacet->dp_byte_count);
8128 if (subfacet->used) {
8129 ds_put_format(&ds, "%.3fs",
8130 (time_msec() - subfacet->used) / 1000.0);
8132 ds_put_format(&ds, "never");
8134 if (subfacet->facet->tcp_flags) {
8135 ds_put_cstr(&ds, ", flags:");
8136 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
8139 ds_put_cstr(&ds, ", actions:");
8140 format_odp_actions(&ds, subfacet->actions, subfacet->actions_len);
8141 ds_put_char(&ds, '\n');
8144 unixctl_command_reply(conn, ds_cstr(&ds));
8149 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
8150 int argc OVS_UNUSED, const char *argv[],
8151 void *aux OVS_UNUSED)
8153 struct ds ds = DS_EMPTY_INITIALIZER;
8154 struct ofproto_dpif *ofproto;
8156 ofproto = ofproto_dpif_lookup(argv[1]);
8158 unixctl_command_reply_error(conn, "no such bridge");
8162 flush(&ofproto->up);
8164 unixctl_command_reply(conn, ds_cstr(&ds));
8169 ofproto_dpif_unixctl_init(void)
8171 static bool registered;
8177 unixctl_command_register(
8179 "bridge {priority tun_id in_port mark packet | odp_flow [-generate]}",
8180 2, 6, ofproto_unixctl_trace, NULL);
8181 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
8182 ofproto_unixctl_fdb_flush, NULL);
8183 unixctl_command_register("fdb/show", "bridge", 1, 1,
8184 ofproto_unixctl_fdb_show, NULL);
8185 unixctl_command_register("ofproto/clog", "", 0, 0,
8186 ofproto_dpif_clog, NULL);
8187 unixctl_command_register("ofproto/unclog", "", 0, 0,
8188 ofproto_dpif_unclog, NULL);
8189 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
8190 ofproto_dpif_self_check, NULL);
8191 unixctl_command_register("dpif/dump-dps", "", 0, 0,
8192 ofproto_unixctl_dpif_dump_dps, NULL);
8193 unixctl_command_register("dpif/show", "[bridge]", 0, INT_MAX,
8194 ofproto_unixctl_dpif_show, NULL);
8195 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
8196 ofproto_unixctl_dpif_dump_flows, NULL);
8197 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
8198 ofproto_unixctl_dpif_del_flows, NULL);
8201 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
8203 * This is deprecated. It is only for compatibility with broken device drivers
8204 * in old versions of Linux that do not properly support VLANs when VLAN
8205 * devices are not used. When broken device drivers are no longer in
8206 * widespread use, we will delete these interfaces. */
8209 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
8211 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
8212 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
8214 if (realdev_ofp_port == ofport->realdev_ofp_port
8215 && vid == ofport->vlandev_vid) {
8219 ofproto->backer->need_revalidate = REV_RECONFIGURE;
8221 if (ofport->realdev_ofp_port) {
8224 if (realdev_ofp_port && ofport->bundle) {
8225 /* vlandevs are enslaved to their realdevs, so they are not allowed to
8226 * themselves be part of a bundle. */
8227 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
8230 ofport->realdev_ofp_port = realdev_ofp_port;
8231 ofport->vlandev_vid = vid;
8233 if (realdev_ofp_port) {
8234 vsp_add(ofport, realdev_ofp_port, vid);
8241 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
8243 return hash_2words(realdev_ofp_port, vid);
8246 /* Returns the ODP port number of the Linux VLAN device that corresponds to
8247 * 'vlan_tci' on the network device with port number 'realdev_odp_port' in
8248 * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9,
8249 * it would return the port number of eth0.9.
8251 * Unless VLAN splinters are enabled for port 'realdev_odp_port', this
8252 * function just returns its 'realdev_odp_port' argument. */
8254 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
8255 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
8257 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
8258 uint16_t realdev_ofp_port;
8259 int vid = vlan_tci_to_vid(vlan_tci);
8260 const struct vlan_splinter *vsp;
8262 realdev_ofp_port = odp_port_to_ofp_port(ofproto, realdev_odp_port);
8263 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
8264 hash_realdev_vid(realdev_ofp_port, vid),
8265 &ofproto->realdev_vid_map) {
8266 if (vsp->realdev_ofp_port == realdev_ofp_port
8267 && vsp->vid == vid) {
8268 return ofp_port_to_odp_port(ofproto, vsp->vlandev_ofp_port);
8272 return realdev_odp_port;
8275 static struct vlan_splinter *
8276 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
8278 struct vlan_splinter *vsp;
8280 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
8281 &ofproto->vlandev_map) {
8282 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
8290 /* Returns the OpenFlow port number of the "real" device underlying the Linux
8291 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
8292 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
8293 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
8294 * eth0 and store 9 in '*vid'.
8296 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
8297 * VLAN device. Unless VLAN splinters are enabled, this is what this function
8300 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
8301 uint16_t vlandev_ofp_port, int *vid)
8303 if (!hmap_is_empty(&ofproto->vlandev_map)) {
8304 const struct vlan_splinter *vsp;
8306 vsp = vlandev_find(ofproto, vlandev_ofp_port);
8311 return vsp->realdev_ofp_port;
8317 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
8318 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
8319 * 'flow->in_port' to the "real" device backing the VLAN device, sets
8320 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
8321 * always the case unless VLAN splinters are enabled), returns false without
8322 * making any changes. */
8324 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
8329 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
8334 /* Cause the flow to be processed as if it came in on the real device with
8335 * the VLAN device's VLAN ID. */
8336 flow->in_port = realdev;
8337 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
8342 vsp_remove(struct ofport_dpif *port)
8344 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8345 struct vlan_splinter *vsp;
8347 vsp = vlandev_find(ofproto, port->up.ofp_port);
8349 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
8350 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
8353 port->realdev_ofp_port = 0;
8355 VLOG_ERR("missing vlan device record");
8360 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
8362 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8364 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
8365 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
8366 == realdev_ofp_port)) {
8367 struct vlan_splinter *vsp;
8369 vsp = xmalloc(sizeof *vsp);
8370 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
8371 hash_int(port->up.ofp_port, 0));
8372 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
8373 hash_realdev_vid(realdev_ofp_port, vid));
8374 vsp->realdev_ofp_port = realdev_ofp_port;
8375 vsp->vlandev_ofp_port = port->up.ofp_port;
8378 port->realdev_ofp_port = realdev_ofp_port;
8380 VLOG_ERR("duplicate vlan device record");
8385 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
8387 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
8388 return ofport ? ofport->odp_port : OVSP_NONE;
8391 static struct ofport_dpif *
8392 odp_port_to_ofport(const struct dpif_backer *backer, uint32_t odp_port)
8394 struct ofport_dpif *port;
8396 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node,
8397 hash_int(odp_port, 0),
8398 &backer->odp_to_ofport_map) {
8399 if (port->odp_port == odp_port) {
8408 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
8410 struct ofport_dpif *port;
8412 port = odp_port_to_ofport(ofproto->backer, odp_port);
8413 if (port && &ofproto->up == port->up.ofproto) {
8414 return port->up.ofp_port;
8420 const struct ofproto_class ofproto_dpif_class = {
8455 port_is_lacp_current,
8456 NULL, /* rule_choose_table */
8463 rule_modify_actions,
8474 get_stp_port_status,
8481 is_mirror_output_bundle,
8482 forward_bpdu_changed,
8483 set_mac_table_config,